Skip to main content

About this book

The 4th International Conference on Hybrid Artificial Intelligence Systems (HAIS 2009), as the name suggests, attracted researchers who are involved in developing and applying symbolic and sub-symbolic techniques aimed at the construction of highly robust and reliable problem-solving techniques, and bringing the most relevant achievements in this field. Hybrid intelligent systems have become increasingly po- lar given their capabilities to handle a broad spectrum of real-world complex problems which come with inherent imprecision, uncertainty and vagueness, hi- dimensionality, and nonstationarity. These systems provide us with the opportunity to exploit existing domain knowledge as well as raw data to come up with promising solutions in an effective manner. Being truly multidisciplinary, the series of HAIS conferences offers an interesting research forum to present and discuss the latest th- retical advances and real-world applications in this exciting research field. This volume of Lecture Notes in Artificial Intelligence (LNAI) includes accepted papers presented at HAIS 2009 held at the University of Salamanca, Salamanca, Spain, June 2009. Since its inception, the main aim of the HAIS conferences has been to establish a broad and interdisciplinary forum for hybrid artificial intelligence systems and asso- ated learning paradigms, which are playing increasingly important roles in a large number of application areas.

Table of Contents


Agents and Multi Agents Systems

Agents in Home Care: A Case Study

Home Care is the term used to refer to any kind of care to a person at his own home. This article presents a case study of the HoCa hybrid multiagent architecture aimed at improving of dependent people in their homes. Hoca architecture uses a set of distributed components to provide a solution to the needs of the assisted people and its main components are software agents that interact with the environment through a distributed communications system. This paper describes the hybrid multiagent system in a home care environment and presents the results obtained.

Juan A. Fraile, Dante I. Tapia, Sara Rodríguez, Juan M. Corchado

EP-MAS.Lib: A MAS-Based Evolutionary Program Approach

Evolutionary/Genetic Programs (EPs) are powerful search techniques used to solve combinatorial optimization problems in many disciplines. Unfortunately, depending on the complexity of the problem, they can be very demanding in terms of computational resources. However, advances in Distributed Artificial Intelligence (DAI), Multi-Agent Systems (MAS) to be more specific, could help users to deal with this matter. In this paper we present an approach in which both technologies, EP and MAS, are combined together aiming to reduce the computational requirements, allowing a response within a reasonable period of time. This approach, called EP-MAS.Lib, is focusing on the interaction among agents in the MAS, and emphasizing on the optimization obtained by means of the evolutionary algorithm/technique. For evaluating the EP-MAS.Lib approach, the paper also presents a case study based on a problem related with the configuration of a neural network for a specific purpose.

Mauricio Paletta, Pilar Herrero

A Framework for Dynamical Intention in Hybrid Navigating Agents

As a foundation for goal-directed behavior, the reactive and deliberative systems of a hybrid agent can share a single, unifying representation of intention. In this paper, we present a framework for incorporating

dynamical intention

into hybrid agents, based on ideas from

spreading activation

models and




) models. In this framework, intentions and other cognitive elements are represented as continuously varying quantities, employed by both sub-deliberative and deliberative processes: On the reactive level, representations support some real-time responsive task re-sequencing; on the deliberative level, representations support common logical reasoning. Because cognitive representations are shared across both levels, inter-level integration is straightforward. Furthermore, dynamical intention is demonstrably consistent with philosophical observations that inform conventional BDI models, so dynamical intentions function as conventional intentions. After describing our framework, we briefly summarize simple demonstrations of our approach, suggesting that dynamical intention-guided intelligence can potentially extend benefits of reactivity without compromising advantages of deliberation in a hybrid agent.

Eric Aaron, Henny Admoni

Multi-agent Based Personal File Management Using Case Based Reasoning

Computer users have been facing a progressively serious problem, namely, how to efficiently manage computer files so as to not only facilitate themselves to use the files, but also save the scare storage resource. Although there are a lot of file management systems available so far, none of them, to the best of our knowledge, can automatically address the deletion/preservation problem of files. To fill this gap, this study explores the value of artificial intelligence techniques in file management. Specifically, this paper develops an intelligent agent based personal file management system, where Case Based Reasoning (CBR) is employed to guide file deletion and preservation. Through some practical experiments, we validate the effectiveness and efficiency of the developed file management system.

Xiaolong Jin, Jianmin Jiang, Geyong Min

Agent-Based Evolutionary System for Traveling Salesman Problem

Evolutionary algorithms are heuristic techniques based on Darwinian model of evolutionary processes, which can be used to find approximate solutions of optimization and adaptation problems. Agent-based evolutionary algorithms are a result of merging two paradigms: evolutionary algorithms and multi-agent systems. In this paper agent-based evolutionary algorithm for solving well known Traveling Salesman Problem is presented. In the experimental part, results of experiments comparing agent-based evolutionary algorithm and classical genetic algorithm are presented.

Rafał Dreżewski, Piotr Woźniak, Leszek Siwik

A Vehicle Routing Problem Solved by Agents

The main purpose of this study is to find out a good solution to the vehicle routing problem considering heterogeneous vehicles.

This problem tries to solve the generation of paths and the assignment of buses on these routes. The objective of this problem is to minimize the number of vehicles required and to maximize the number of demands transported.

This paper considers a Memetic Algorithm for the vehicle routing problem with heterogeneous fleet for any transport problem between many origins and many destinations. A Memetic Algorithm always maintains a population of different solutions to the problem, each of which operates as an agent. These agents interact between themselves within a framework of competition and cooperation.

Extensive computational tests on some instances taken from the literature reveal the effectiveness of the proposed algorithm.

Ma Belén Vaquerizo García

MACSDE: Multi-Agent Contingency Response System for Dynamic Environments

Dynamic environments represent a quite complex domain, where the information available changes continuously. In this paper, a contingency response system for dynamic environments called MACSDE is presented. The explained system allows the introduction of information, the monitoring of the process and the generation of predictions. The system makes use of a Case-Based Reasoning system which generates predictions using previously gathered information. It employs a distributed multi-agent architecture so that the main components of the system can be accessed remotely. Therefore, all functionalities can communicate in a distributed way, even from mobile devices. The core of the system is a group of deliberative agents acting as controllers and administrators for all functionalities. The system explained includes a novel network for data classification and retrieval. Such network works as a summarization algorithm for the results of an ensemble of Self-Organizing Maps. The presented system has been tested with data related with oil spills and forest fire, obtaining quite hopeful results.

Aitor Mata, Belén Pérez, Angélica González, Bruno Baruque, Emilio Corchado

HAIS Applications

Measuring and Visualising Similarity of Customer Satisfaction Profiles for Different Customer Segments

Questionnaires are a common tool to gain insight to customer satisfaction. The data available from such questionnaires is an important source of information for a company to judge and improve its performance in order to achieve maximum customer satisfaction. Here, we are interested in finding out, how much individual customer segments are similar or differ w.r.t. to their satisfaction profiles. We propose a hybrid approach using measures for the similarity of satisfaction profiles based on principles from statistics in combination with visualization techniques. The applicability and benefit of our approach is demonstrated on the basis of real-world customer data.

Frank Klawonn, Detlef D. Nauck, Katharina Tschumitschew

Development of a Decision-Maker in an Anticipatory Reasoning-Reacting System for Terminal Radar Control

Terminal radar control is more and more complex in recent years. To reduce human errors in terminal radar control, an automatic system to support conflict detection and conflict resolution is required for reliable and safe terminal radar control. An anticipatory reasoning-reacting system for terminal radar control is a hopeful candidate for such systems. This paper proposes a methodology of decision-making in an anticipatory reasoning-reacting system for terminal radar control, presents a prototype of decision-maker, and shows that it can make appropriate decisions in anticipatory reasoning-reacting system for terminal radar control.

Natsumi Kitajima, Yuichi Goto, Jingde Cheng

Study of Outgoing Longwave Radiation Anomalies Associated with Two Earthquakes in China Using Wavelet Maxima

The paper presents an analysis of the continuous outgoing longwave radiation (OLR) based on time and space by using the wavelet-based data mining techniques. The analyzed results reveal that the anomalous variations exist prior to the earthquakes. The methods studied in this work include wavelet transformations and spatial/temporal continuity analysis of wavelet maxima. These methods have been applied to detect singularities from OLR data that correspond to seismic precursors, particularly to a comparative study of the two earthquakes of Wenchuan and Pure recently occurred in China.

Pan Xiong, Yaxin Bi, Xuhui Shen

A Hybrid Approach for Designing the Control System for Underwater Vehicles

An approach in the form of an automatic evolutionary design environment for obtaining any type of control systems for underwater vehicles is presented. A specific case is considered in which this strategy is hybridized with Artificial Neural Networks. The design procedure is carried out by means of evolutionary techniques from a set of specifications using as a fitness evaluator an ad-hoc hydrodynamic simulator which includes the estimation of added mass and added inertia coefficients. The resulting design environment was used to construct the neural network based controllers of a submersible catamaran. Results of the application of the automatic design procedure and of the operation of the controllers thus obtained are presented.

A. Lamas, F. López Peña, R. J. Duro

Hydrodynamic Design of Control Surfaces for Ships Using a MOEA with Neuronal Correction

In this paper we present a hybrid intelligent system for the hydrodynamic design of control surfaces on ships. Our main contribution here is the hybridization of Multiobjective Evolutionary Algorithms (MOEA) and a neural correction procedure in the fitness evaluation stage that permits obtaining solutions that are precise enough for the MOEA to operate with, while drastically reducing the computational cost of the simulation stage for each individual. The MOEA searches for the optimal solutions and the neuronal system corrects the deviations of the simplified simulation model to obtain a more realistic design. This way, we can exploit the benefits of a MOEA decreasing the computational cost in the evaluation of the candidate solutions while preesrving the reliability of the simulation model. The proposed hybrid system is successfully applied in the design of a 2D control surface for ships and extended to a 3D one.

V. Díaz-Casás, Francisco Bellas, Fernando López-Peña, Richard Duro

Closures of Downward Closed Representations of Frequent Patterns

The discovery of frequent patterns is one of the most important issues in the data mining area. A major difficulty concerning frequent patterns is huge amount of discovered patterns. The problem can be significantly alleviated by applying concise representations of frequent patterns. In this paper, we offer new lossless representations of frequent patterns that are derivable from downward closed representations by replacing the original elements and eventually some border ones with their closures. We show for which type of downward closed representations the additional closures are superfluous and for which they need to be stored. If the additional closures are not stored, the new representations are guaranteed to be not less concise than the original ones.

Marzena Kryszkiewicz

Transductive-Weighted Neuro-Fuzzy Inference System for Tool Wear Prediction in a Turning Process

This paper presents the application to the modeling of a novel technique of artificial intelligence. Through a transductive learning process, a neuro-fuzzy inference system enables to create a different model for each input to the system at issue. The model was created from a given number of known data with similar features to data input. The sum of these individual models yields greater accuracy to the general model because it takes into account the particularities of each input. To demonstrate the benefits of this kind of modeling, this system is applied to the tool wear modeling for turning process.

Agustín Gajate, Rodolfo E. Haber, José R. Alique, Pastora I. Vega

Review of Hybridizations of Kalman Filters with Fuzzy and Neural Computing for Mobile Robot Navigation

Kalman Filters (KF) are at the root of many computational solutions for autonomous systems navigation problems, besides other application domains. The basic linear formulation has been extended in several ways to cope with non-linar dynamic environments. One of the latest trend is to introduce other Computational Intelligence (CI) tools, such as Fuzzy Systems or Artificial Neural Networks inside its computational loop, in order to obtain learning and advanced adaptive properties. This paper offers a short review of current approaches.

Manuel Graña, Iván Villaverde, Jose Manuel López Guede, Borja Fernández

A Real-Time Person Detection Method for Moving Cameras

In this paper, we introduce an advanced real-time method for vision-based pedestrian detection made up by the sequential combination of two basic methods applied in a coarse to fine fashion. The proposed method aims to achieve an improved balance between detection accuracy and computational load by taking advantage of the strengths of these basic techniques. Boosting techniques in human detection, which have been demonstrated to provide rapid but not accurate enough results, are used in the first stage to provide a preliminary candidate selection in the scene. Then, feature extraction and classification methods, which present high accuracy rates at expenses of a higher computational cost, are applied over boosting candidates providing the final prediction. Experimental results show that the proposed method performs effectively and efficiently, which supports its suitability for real applications.

Javier Oliver, Alberto Albiol, Samuel Morillas, Guillermo Peris-Fajarnés

Unsupervised Methods for Anomalies Detection through Intelligent Monitoring Systems

The success of intelligent diagnosis systems normally depends on the knowledge about the failures present on monitored systems. This knowledge can be modelled in several ways, such as by means of rules or probabilistic models. These models are validated by checking the system output fit to the input in a supervised way. However, when there is no such knowledge or when it is hard to obtain a model of it, it is alternatively possible to use an unsupervised method to detect anomalies and failures. Different unsupervised methods (HCL, K-Means, SOM) have been used in present work to identify abnormal behaviours on the system being monitored. This approach has been tested into a real-world monitored system related to the railway domain, and the results show how it is possible to successfully identify new abnormal system behaviours beyond those previously modelled well-known problems.

Alberto Carrascal, Alberto Díez, Ander Azpeitia

Architecture for Hybrid Robotic Behavior

Software architectures for agent technology and robots have been polarized between reactive architectures and architectures based on planning and reasoning. Although hybrid architectures have been shown to offer benefits from both, these seem complicated to integrate. In this paper we integrate the reactive nature of finite state machines and the reasoning capabilities of non-monotonic logics to produce intelligent autonomous robots. In particular, we demonstrate this with a robotic poker player. The robotic player integrates vision, sound recognition, motion control and the reasoning to perform competitively as a player in a complex game with incomplete information.

David Billington, Vladimir Estivill-Castro, René Hexel, Andrew Rock

A Hybrid Solution for Advice in the Knowledge Management Field

This paper presents a hybrid artificial intelligent solution that helps to automatically generate proposals, aimed at improving the internal states of organization units from a Knowledge Management (KM) point of view. This solution is based on the combination of the Case-Based Reasoning (CBR) and connectionist paradigms. The required outcome consists of customized solutions for different areas of expertise related to the organization units, once a lack of knowledge in any of those has been identified. On the other hand, the system is fed with KM data collected at the organization and unit contexts. This solution has been integrated in a KM system that additionally profiles the KM status of the whole organization.

Álvaro Herrero, Aitor Mata, Emilio Corchado, Lourdes Sáiz

Cluster Analysis

A Cluster-Based Feature Selection Approach

This paper proposes a filter-based method for feature selection. The filter is based on the partitioning of the feature space into clusters of similar features. The number of clusters and, consequently, the cardinality of the subset of selected features, is automatically estimated from the data. Empirical results illustrate the performance of the proposed algorithm, which in general has obtained competitive results in terms of classification accuracy when compared to a state of the art algorithm for feature selection, but with more modest computing time requirements.

Thiago F. Covões, Eduardo R. Hruschka, Leandro N. de Castro, Átila M. Santos

Automatic Clustering Using a Synergy of Genetic Algorithm and Multi-objective Differential Evolution

This paper applies the Differential Evolution (DE) and Genetic Algorithm (GA) to the task of automatic fuzzy clustering in a Multi-objective Optimization (MO) framework. It compares the performance a hybrid of the GA and DE (GADE) algorithms over the fuzzy clustering problem, where two conflicting fuzzy validity indices are simultaneously optimized. The resultant Pareto optimal set of solutions from each algorithm consists of a number of non-dominated solutions, from which the user can choose the most promising ones according to the problem specifications. A real-coded representation of the search variables, accommodating variable number of cluster centers, is used for GADE. The performance of GADE has also been contrasted to that of two most well-known schemes of MO.

Debarati Kundu, Kaushik Suresh, Sayan Ghosh, Swagatam Das, Ajith Abraham, Youakim Badr

Credibility Coefficients in Hybrid Artificial Intelligence Systems

ARES System is an application dedicated to data analysis supported by Rough Set theory. Currently the system is expanded by such approaches as Emerging Patterns and Support Vector Machine. A unique feature of ARES System is applying credibility coefficients to identify improper objects within information systems. The credibility coefficient is a measure, which attempts to assess a degree of typicality of each object in respect to the rest of information system. The paper presents a concept of credibility coefficients in context of hybrid artificial intelligence systems combined on ARES System platform. Ordinal credibility coefficient supports aggregation of number incomparable credibility coefficients based on different approaches.

Roman Podraza

An Evolutionary Algorithm for Missing Values Substitution in Classification Tasks

This paper proposes a method for substituting missing values that is based on an evolutionary algorithm for clustering. Missing values substitution has been traditionally assessed by some measures of the prediction capability of imputation methods. Although this evaluation is useful, it does not allow inferring the influence of imputed values in the ultimate modeling task (e.g., in classification). In this sense, alternative approaches to the so called prediction capability evaluation are needed. Therefore, we here also discuss the influence of imputed values in the classification task. Preliminary results obtained in a bioinformatics data set illustrate that the proposed imputation algorithm can insert less classification bias than three state of the art algorithms (i.e., KNNimpute, SKNN and IKNN). Finally, we illustrate that better prediction results do not necessarily imply in less classification bias.

Jonathan de A. Silva, Eduardo R. Hruschka

Data Mining and Knowledge Discovery

A Generic and Extendible Multi-Agent Data Mining Framework

A generic and extendible Multi-Agent Data Mining (MADM) framework, EMADS (the Extendible Multi-Agent Data mining System) is described. The central feature of the framework is that it avoids the use of agreed meta-language formats by supporting a system of wrappers. The advantage offered is that the system is easily extendible, so that further data agents and mining agents can simply be added to the system. A demonstration EMADS framework is currently available. The paper includes details of the EMADS architecture and the wrapper principle incorporated into it. A full description and evaluation of the framework’s operation is provided by considering two MADM scenarios.

Kamal Ali Albashiri, Frans Coenen

A Modular Distributed Decision Support System with Data Mining Capabilities

Although Decision Support Systems (DSSs) to help control strategies have been developed and improved for about two decades, their technology is still limited to end-to-end solutions. This paper proposes a framework for developing a Modular Distributed Decision Support System (MDDSS), capable of handling global knowledge expressed in various forms and accessible by different users with different needs. The interaction of human experts in different parts of the world and intelligent distributed modules will allow the system to deal with increased volume of knowledge (impossible with existing systems) and also to be easily upgradable to future technologies.

Leonardo Gualano, Paul Young

A Fuzzy Quantitative Integrated Metric Model for CMMI Appraisal

In Capability Maturity Model Integrated (CMMI) systems, the Lead Appraiser evaluates the processes of one company according to the qualitative methods, such as questionnaire, interview, and document. A Fuzzy Quantitative Integrated Metric Model (FQIMM) is proposed in this paper due to the subjective measurement and non-quantitative representations of Lead Appraiser. The FQIMM integrates Quantitative Software Metrics Set, linguistic variables and interval of confidence. FQIMM can help software development companies to evaluate competitiveness by quantitative approach and then know their position more quickly and effectively.

Ching-Hsue Cheng, Jing-Rong Chang, Chen-Yi Kuo, Shu-Ying Liao

Analyzing Transitive Rules on a Hybrid Concept Discovery System

Multi-relational concept discovery aims to find the relational rules that best describe the target concept. An important challenge that relational knowledge discovery systems face is intractably large search space and there is a trade-off between pruning the search space for fast discovery and generating high quality rules. Combining ILP approach with conventional association rule mining techniques provides effective pruning mechanisms. Due to the nature of Apriori algorithm, the facts that do not have common attributes with the target concept are discarded. This leads to efficient pruning of search space. However, under certain conditions, it fails to generate transitive rules, which is an important drawback when transitive rules are the only way to describe the target concept. In this work, we analyze the effect of incorporating unrelated facts for generating transitive rules in an hybrid relational concept discovery system, namely C


D, which combines ILP and Apriori.

Yusuf Kavurucu, Pinar Senkul, Ismail Hakki Toroslu

Survey of Business Intelligence for Energy Markets

Today, there is the need for establishing a strong relationship between Business Intelligence (BI) and Energy Markets (EM). This is crucial because of enormous and increasing data volumes generated and stored day by day in the EM. The data volume turns impossible to obtain clear data understanding through human analysis or with traditional tools. BI can be the solution. In this sense, we present a comprehensive survey related with the BI applications for the EM, in order to show trends and useful methods for tackling down every day EM challenges. We outline how BI approach can effectively support a variety of difficult and challenging EM issues like prediction, pattern recognition, modeling and others. We can observe that hybrid artificial intelligence systems are common in EM. An extensive bibliography is also included.

Manuel Mejía-Lavalle, Ricardo Sosa R., Nemorio González M., Liliana Argotte R.

Evolutionary Computation

Hybrid Multilogistic Regression by Means of Evolutionary Radial Basis Functions: Application to Precision Agriculture

In this paper, a previously defined hybrid multilogistic regression model is extended and applied to a precision agriculture problem. This model is based on a prediction function which is a combination of the initial covariates of the problem and the hidden neurons of an Artificial Neural Network (ANN). Several statistical and soft computing techniques have been applied for determining these models such as logistic regression, ANNs and Evolutionary Algorithms (EAs). This paper proposes the use of Radial Basis Functions (RBFs) transformations for this model. The estimation of the coefficients of the model is basically carried out in two phases. First, the number of RBFs and the radii and centers’ vector are determined by means of an EA. Afterwards, the new RBF nonlinear transformations obtained for the best individual in the last generation are added to the covariate space. Finally, a maximum likelihood optimization method determines the rest of the coefficients of the multilogistic regression model. In order to determine the performance of this approach, it has been applied to a problem of discriminating cover crops in olive orchards affected by its phenological stage using their spectral signatures obtained with a high-resolution field spectroradiometer. The empirical results for this complex real agronomical problem and the corresponding Dunnet statistical test carried out show that the proposed model is very promising in terms of classification accuracy and number of wavelengths used by the classifier.

P. A. Gutiérrez, C. Hervás-Martínez, J. C. Fernández, F. López-Granados

Economic Load Dispatch Using a Chemotactic Differential Evolution Algorithm

This paper presents a novel stochastic optimization approach to solve constrained economic load dispatch (ELD) problem using Hybrid Bacterial Foraging-Differential Evolution optimization algorithm. In this hybrid approach computational chemotaxis of BFOA, which may also be viewed as a stochastic gradient search, has been coupled with DE type mutation and crossover of the optimization agents. The proposed methodology easily takes care of solving non-convex economic load dispatch problems along with different constraints like transmission losses, dynamic operation constraints (ramp rate limits) and prohibited operating zones. Simulations were performed over various standard test systems with different number of generating units and comparisons are performed with other existing relevant approaches. The findings affirmed the robustness and proficiency of the proposed methodology over other existing techniques.

Arijit Biswas, Sambarta Dasgupta, Bijaya K. Panigrahi, V. Ravikumar Pandi, Swagatam Das, Ajith Abraham, Youakim Badr

Cellular Automata Rule Detection Using Circular Asynchronous Evolutionary Search

A circular evolutionary model is proposed to produce

Cellular Automata (CA)

rules for the computationally emergent task of density classification. The task refers to determining the initial density most present in the initial cellular state of a one-dimensional cellular automaton within a number of update steps. This is a challenging problem extensively studied due to its simplicity and potential to generate a variety of complex behaviors. The proposed circular evolutionary model aims to facilitate a good exploitation of relevant genetic material while increasing the population diversity. This goal is achieved by integrating a fitness guided population topology with an asynchronous search scheme. Both selection and recombination take place asynchronously enabling a gradual propagation of information from the fittest individuals towards the less fit members of the population. Numerical experiments emphasize a competitive performance of the circular search algorithm compared to other evolutionary models indicating the potential of the proposed model.

Anca Gog, Camelia Chira

Evolutionary Non-linear Great Deluge for University Course Timetabling

This paper presents a hybrid evolutionary algorithm to tackle university course timetabling problems. The proposed approach is an extension of a non-linear great deluge algorithm in which evolutionary operators are incorporated. First, we generate a population of feasible solutions using a tailored process that incorporates heuristics for graph colouring and assignment problems. That initialisation process is capable of producing feasible solutions even for the large and most constrained problem instances. Then, the population of feasible timetables is subject to a steady-state evolutionary process that combines mutation and stochastic local search. We conduct experiments to evaluate the performance of the proposed hybrid algorithm and in particular, the contribution of the evolutionary operators. Our results show that the hybrid between non-linear great deluge and evolutionary operators produces very good results on the instances of the university course timetabling problem tackled here.

Dario Landa-Silva, Joe Henry Obit

Co-operative Co-evolutionary Approach to Multi-objective Optimization

Co-evolutionary algorithms are evolutionary algorithms in which the given individual’s fitness value estimation is made on the basis of interactions of this individual with other individuals present in the population. In this paper agent-based versions of co-operative co-evolutionary algorithms are presented and evaluated with the use of standard multi-objective test functions. The results of experiments are used to compare proposed agent-based co-evolutionary algorithms with state-of-the-art multi-objective evolutionary algorithms: SPEA2 and NSGA-II.

Rafał Dreżewski, Krystian Obrocki

A GA(TS) Hybrid Algorithm for Scheduling in Computational Grids

The hybridization of heuristics methods aims at exploring the synergies among stand alone heuristics in order to achieve better results for the optimization problem under study. In this paper we present a hybridization of Genetic Algorithms (GAs) and Tabu Search (TS) for scheduling in computational grids. The purpose in this hybridization is to benefit the exploration of the solution space by a population of individuals with the exploitation of solutions through a smart search of the TS. Our GA(TS) hybrid algorithm runs the GA as the main algorithm and calls TS procedure to improve individuals of the population. We evaluated the proposed hybrid algorithm using different Grid scenarios generated by a Grid simulator. The computational results showed that the hybrid algorithm outperforms both the GA and TS for the makespan value but cannot outperform them for the flowtime of the scheduling.

Fatos Xhafa, Juan A. Gonzalez, Keshav P. Dahal, Ajith Abraham

On the Model–Building Issue of Multi–Objective Estimation of Distribution Algorithms

It has been claimed that perhaps a paradigm shift is necessary in order to be able to deal with this scalability issue of multi–objective optimization evolutionary algorithms. Estimation of distribution algorithms are viable candidates for such task because of their adaptation and learning abilities and simplified algorithmics. Nevertheless, the extension of EDAs to the multi–objective domain have not provided a significant improvement over MOEAs.

In this paper we analyze the possible causes of this underachievement and propose a set of measures that should be taken in order to overcome the current situation.

Luis Martí, Jesús García, Antonio Berlanga, José M. Molina

A Hooke-Jeeves Based Memetic Algorithm for Solving Dynamic Optimisation Problems

Dynamic optimisation problems are difficult to solve because they involve variables that change over time. In this paper, we present a new Hooke-Jeeves based Memetic Algorithm (HJMA) for dynamic function optimisation, and use the Moving Peaks (MP) problem as a test bed for experimentation. The results show that HJMA outperforms all previously published approaches on the three standardised benchmark scenarios of the MP problem. Some observations on the behaviour of the algorithm suggest that the original Hooke-Jeeves algorithm is surprisingly similar to the simple local search employed for this task in previous work.

Irene Moser, Raymond Chiong

Hybrid Evolutionary Algorithm for Solving Global Optimization Problems

Differential Evolution (DE) is a novel evolutionary approach capable of handling non-differentiable, non-linear and multi-modal objective functions. DE has been consistently ranked as one of the best search algorithm for solving global optimization problems in several case studies. This paper presents a simple and modified hybridized Differential Evolution algorithm for solving global optimization problems. The proposed algorithm is a hybrid of Differential Evolution (DE) and Evolutionary Programming (EP). Based on the generation of initial population, three versions are proposed. Besides using the uniform distribution (U-MDE), the Gaussian distribution (G-MDE) and Sobol sequence (S-MDE) are also used for generating the initial population. Empirical results show that the proposed versions are quite competent for solving the considered test functions.

Radha Thangaraj, Millie Pant, Ajith Abraham, Youakim Badr

Learning Algorithms

Fragmentary Synchronization in Chaotic Neural Network and Data Mining

This paper proposes an improved model of chaotic neural network used to cluster high-dimensional datasets with cross sections in the feature space. A thorough study was designed to elucidate the possible behavior of hundreds interacting chaotic oscillators. New synchronization type - fragmentary synchronization within cluster elements dynamics was found. The paper describes a method for detecting fragmentary synchronization and it’s advantages when applied to data mining problem.

Elena N. Benderskaya, Sofya V. Zhukova

Two-Stage Neural Network Approach to Precise 24-Hour Load Pattern Prediction

The paper presents the neural network approach to the precise 24-hour load pattern prediction for the next day in the power system. In this approach we use the ensemble of few neural network predictors working in parallel. The predicted series containing 24 values of the load pattern generated by the neural predictors are combined together using principal component analysis. Few principal components form the input vector for the final stage predictor composed of another neural network. The developed system of prediction was tested on the real data of the Polish Power System. The results have been compared to the appropriate values generated by other methods.

Krzysztof Siwek, Stanislaw Osowski

Tentative Exploration on Reinforcement Learning Algorithms for Stochastic Rewards

This paper addresses a way to generate mixed strategies using reinforcement learning algorithms in domains with stochastic rewards. A new algorithm, based on Q-learning model, called TERSQ is introduced. As a difference from other approaches for stochastic scenarios, TERSQ uses a global exploration rate for all the state/actions in the same run. This exploration rate is selected at the beginning of each round, using a probabilistic distribution, which is updated once the run is finished. In this paper we compare TERSQ with similar approaches that use probability distributions depending on state-action pairs. Two experimental scenarios have been considered. First one deals with the problem of learning the optimal way to combine several evolutionary algorithms used simultaneously by a hybrid approach. In the second one, the objective is to learn the best strategy for a set of competing agents in combat-based videogame.

Luis Peña, Antonio LaTorre, José-María Peña, Sascha Ossowski

Comparative Evaluation of Semi-supervised Geodesic GTM

In many real problems that ultimately require data classification, not all the class labels are readily available. This concerns the field of semi-supervised learning, in which missing class labels must be inferred from the available ones as well as from the natural cluster structure of the data. This structure can sometimes be quite convoluted. Previous research has shown the advantage, for these cases, of using the geodesic metric in clustering models of the manifold learning family to reveal the underlying true data structure. In this brief paper, we present a novel semi-supervised approach, namely Semi-Supervised Geo-GTM (SS-Geo-GTM). This is an extension of Geo-GTM, a variation on the Generative Topographic Mapping (GTM) manifold learning model for data clustering and visualization that resorts to the geodesic metric. SS-Geo-GTM uses a proximity graph built from Geo-GTM manifold as the basis for a label propagation algorithm that infers missing class labels. Its performance is compared to those of a semi-supervised version of the standard GTM and of the alternative Laplacian Eigenmaps method.

Raúl Cruz-Barbosa, Alfredo Vellido

Special Session

Real World HAIS Applications and Data Uncertainty

Application of Interval Type-2 Fuzzy Logic Systems for Control of the Coiling Entry Temperature in a Hot Strip Mill

An interval type-2 fuzzy logic system is used to setup the cooling water applied to the strip as it traverses the run out table in order to achieve the coiler entry temperature target. The interval type-2 fuzzy setup model uses as inputs the target coiling entry temperature, the target strip thickness, the predicted finish mill exit temperature and the target finishing mill exit speed. The experimental results of the application of the interval type-2 fuzzy logic system for coiler entry temperature prediction in a real hot strip mill were carried out for three different types of coils. They proved the feasibility of the systems developed here for coiler entry temperature prediction. Comparison with an on-line type-1 fuzzy logic based model shows that the interval type-2 fuzzy logic system improves performance in coiler entry temperature prediction under the tested condition.

Gerardo M. Méndez, Luis Leduc-Lezama, Rafael Colas, Gabriel Murillo-Pérez, Jorge Ramírez-Cuellar, José J. López

A Review on the Application of Hybrid Artificial Intelligence Systems to Optimization Problems in Operations Management

The use of hybrid artificial intelligence systems in operations management has grown during the last years given their ability to tackle combinatorial and NP hard problems. Furthermore, operations management problems usually involve imprecision, uncertainty, vagueness, and high-dimensionality. This paper examines recent developments in the field of hybrid artificial intelligence systems for those operations management problems where hybrid approaches are more representative: design engineering, process planning, assembly line balancing, and dynamic scheduling.

Oscar Ibáñez, Oscar Cordón, Sergio Damas, Luis Magdalena

A Pool of Experts to Evaluate the Evolution of Biological Processes in SBR Plants

In order to minimize the costs and maximize its efficiency, a Sequencing Batch Reactor requires continuous monitoring. The sensors it can be reasonably equipped with provide only indirect information on the state of the chemical reactions taking place in the tank, so the data must be analysed and interpreted. At present, no optimum, completely reliable procedure exists: instead, there exist several criteria which can be applied under different conditions. This paper shows that estimating the confidence in the quality of the response of a criterion can increase the robustness of a criterion. Then, interpreting the responses in terms of possibility distributions, the different answers can be merged, thus obtaining a more reliable overall estimate.

Davide Sottara, Gabriele Colombini, Luca Luccarini, Paola Mello

A Hybrid Ant-Based Approach to the Economic Triangulation Problem for Input-Output Tables


Triangulation Problem for Input-Output Matrices

has been intensively studied in order to understand the complex series of interactions among the sectors of an economy. The problem refers to finding a simultaneously permutation of rows and columns of a matrix such as the sum of the entries which are above the main diagonal is maximum. This is a linear ordering problem – a well-known NP-hard combinatorial optimization problem. A new hybrid heuristic based on ant algorithms is proposed to efficiently solve the triangulation problem. Starting from a greedy solution, the proposed model hybridizes the

Ant Colony System (ACS)

metaheuristic with an

Insert-Move (IM)

local search mechanism able to refine ant solutions. The proposed


algorithm is tested with good results on some real-life economic data sets.

Camelia-M. Pintea, Gloria Cerasela Crisan, Camelia Chira, D. Dumitrescu

A Thermodynamical Model Study for an Energy Saving Algorithm

A local Spanish company that produces electric heaters needs an energy saving device to be integrated with the heaters. It was proven that a hybrid artificial intelligent systems (HAIS) could afford the energy saving reasonably, even though some improvements must be introduced. One of the critical elements in the process of designing an energy saving system is the thermodynamical modeling of the house to be controlled. This work presents a study of different first order techniques, some taken from the literature and other new proposals, for the prediction of the thermal dynamics in a house. Finally it is concluded that a first order prediction system is not a valid prediction model for such an energy saving system.

Enrique de la Cal, José Ramón Villar, Javier Sedano

Applications of Hybrid Artificial Intelligence in Bioinformatics

A Fuzzy Approach of the Kohonen’s Maps Applied to the Analysis of Biomedical Signals

Self-organizing maps have been used successfully in pattern classification problems related to many areas of knowledge and also applied as a tool for statistical multivariate data analysis. Data classification via self-organizing maps deals specifically with relations between objects, meaning that there are limitations to define class limits when an object belonging to a particular class “migrates” to another one. To address this issue, a solution involving self-organizing maps and fuzzy logic is proposed with the objective of generating a neighborhood between these classes. The developed system receives the network output and automatically generates self-organizing maps. This unified vision of the model is used in the analyzing biomedical signals in diabetic patients for monitoring blood glucose stage. Early diagnosis and glucose signals monitoring can prevent or delay the initiation and development of clinical complications related to diabetes.

Andrilene Maciel, Luis Coradine, Roberta Vieira, Manoel Lima

Unearth the Hidden Supportive Information for an Intelligent Medical Diagnostic System

This paper presents an intelligent diagnostic supporting system –



DiaKAW (Intelligent and Interactive Diagnostic Knowledge Acquisition Workbench), which automatically extracts useful knowledge from massive medical data to support real medical diagnosis. In which, our two novel pre-processing algorithms MIDCA (Multivariate Interdependent Discretization for Continuous-valued Attributes) and LUIFS (Latent Utility of Irrelevant Feature Selection) for continuous feature discretization (CFD) and feature selection (FS) respectively, assist in accelerating the diagnostic accuracy by taking the attributes’ supportive relevance into consideration during the data preparation process. Such strategy minimizes the information lost and maximizes the intelligence and accuracy of the system. The empirical results on several real-life datasets from UCI repository demonstrate the goodness of our diagnostic system.

Sam Chao, Fai Wong

Incremental Kernel Machines for Protein Remote Homology Detection

Protein membership prediction is a fundamental task to retrieve information for unknown or unidentified sequences. When support vector machines (SVMs) are associated with the right kernels, this machine learning technique can build state-of-the-art classifiers. However, traditional implementations work in a batch fashion, limiting the application to very large and high dimensional data sets, typical in biology. Incremental SVMs introduce an alternative to batch algorithms, and a good candidate to solve these problems. In this work several experiments are conducted to evaluate the performance of the incremental SVM on remote homology detection using a benchmark data set. The main advantages are shown, opening the possibility to further improve the algorithm in order to achieve even better classifiers.

Lionel Morgado, Carlos Pereira

Use of Classification Algorithms in Noise Detection and Elimination

Data sets in Bioinformatics usually present a high level of noise. Various processes involved in biological data collection and preparation may be responsible for the introduction of this noise, such as the imprecision inherent to laboratory experiments generating these data. Using noisy data in the induction of classifiers through Machine Learning techniques may harm the classifiers prediction performance. Therefore, the predictions of these classifiers may be used for guiding noise detection and removal. This work compares three approaches for the elimination of noisy data from Bioinformatics data sets using Machine Learning classifiers: the first is based in the removal of the detected noisy examples, the second tries to reclassify these data and the third technique, named hybrid, unifies the previous approaches.

André L. B. Miranda, Luís Paulo F. Garcia, André C. P. L. F. Carvalho, Ana C. Lorena

SGNG Protein Classifier by Matching 3D Structures

In this paper, a novel 3D structure-based approach is presented for fast and accurate classification of protein molecules. We have used our voxel and ray based descriptors for feature extraction of protein structures. By using these descriptors, in this paper we propose a novel approach for classifying protein molecules, named Supervised Growing Neural Gas (SGNG). It combines the Growing Neural Gas (GNG) as a hidden layer, and Radial Basis Function (RBF) as an output layer. GNG and its supervised version SGNG have not yet been applied for protein retrieval and classification. Our approach was evaluated according to the SCOP method. The results show that our approach achieves more than 83,5% by using the voxel descriptor and 98,4% classification accuracy by using the ray descriptor, while it is simpler and faster than the SCOP method. We provide some experimental results.

Georgina Mirceva, Andrea Kulakov, Danco Davcev

Evolutionary Multiobjective Machine Learning

Memetic Pareto Differential Evolution for Designing Artificial Neural Networks in Multiclassification Problems Using Cross-Entropy Versus Sensitivity

This work proposes a Multiobjective Differential Evolution algorithm based on dominance Pareto concept for multiclassification problems using multilayer perceptron neural network models. The algorithm include a local search procedure and optimizes two conflicting objectives of multiclassifiers, a high correct classification rate and a high classification rate for each class, of which the latter is not usually optimized in classification. Once the Pareto front is built, we use two automatic selection methodologies of individuals: the best model with respect to accuracy and the best model with respect to sensitivity (extremes in the Pareto front). These strategies are applied to solve six classification benchmark problems obtained from the UCI repository. The models obtained show a high accuracy and a high classification rate for each class.

Juan Carlos Fernández, César Hervás, Francisco José Martínez, Pedro Antonio Gutiérrez, Manuel Cruz

Pareto-Based Multi-output Model Type Selection

In engineering design the use of approximation models (= surrogate models) has become standard practice for design space exploration, sensitivity analysis, visualization and optimization. Popular surrogate model types include neural networks, support vector machines, Kriging models, and splines. An engineering simulation typically involves multiple response variables that must be approximated. With many approximation methods available, the question of which method to use for which response consistently arises among engineers and domain experts. Traditionally, the different responses are modeled separately by independent models, possibly involving a comparison among model types. Instead, this paper proposes a multi-objective approach can benefit the domain expert since it enables automatic model


selection for each output on the fly without resorting to multiple runs. In effect the optimal model complexity and model type for each output is determined automatically. In addition a multi-objective approach gives information about output correlation and facilitates the generation of diverse ensembles. The merit of this approach is illustrated with a modeling problem from aerospace.

Dirk Gorissen, Ivo Couckuyt, Karel Crombecq, Tom Dhaene

A Comparison of Multi-objective Grammar-Guided Genetic Programming Methods to Multiple Instance Learning

This paper develops a first comparative study of multi- objective algorithms in Multiple Instance Learning (MIL) applications. These algorithms use grammar-guided genetic programming, a robust classification paradigm which is able to generate understandable rules that are adapted to work with the MIL framework. The algorithms obtained are based on the most widely used and compared multi-objective evolutionary algorithms. Thus, we design and implement SPG3P-MI based on the Strength Pareto Evolutionary Algorithm, NSG3P-MI based on the Non-dominated Sorting Genetic Algorithm and MOGLG3P-MI based on the Multi-objective genetic local search. These approaches are tested with different MIL applications and compared to a previous single-objective grammar-guided genetic programming proposal. The results demonstrate the excellent performance of multi-objective approaches in achieving accurate models and their ability to generate comprehensive rules in the knowledgable discovery process.

Amelia Zafra, Sebastián Ventura

Hybrid Reasoning and Coordination Methods on Multi-Agent Systems

On the Formalization of an Argumentation System for Software Agents

Argumentation techniques for multi-agent systems (MAS) coordination are relatively common nowadays. But most frameworks are theoretical approaches to the problem. ASBO is an Argumentation System Based on Ontologies. It follows an engineering oriented approach to materialize a software tool which allows working with argumentation in MAS. But ASBO has also a formal model in the background. This paper introduces such formal model, as a way to identify and unambiguously define the core elements that argumentation systems should include.

Andres Munoz, Juan A. Botia

A Dialogue-Game Approach for Norm-Based MAS Coordination

Open societies are situated in dynamic environments and are formed by heterogeneous autonomous agents. For ensuring social order, norms have been employed as coordination mechanisms. However, the dynamical features of open systems may cause that norms loose their validity and need to be adapted. Therefore, this paper proposes a new dialogue game protocol for modelling the interactions produced between agents that must reach an agreement on the use of norms. An application example has been presented for showing the performance of the protocol and its usefulness as a mechanism for managing the solving process of a coordination problem through norms.

S. Heras, N. Criado, E. Argente, V. Julián

Incorporating a Temporal Bounded Execution to the CBR Methodology

In real-time Multi-Agent Systems, Real-Time Agents merge intelligent deliberative techniques with real-time reactive actions in a distributed environment. CBR has been successfully applied in Multi-Agent Systems as deliberative mechanism for agents. However, in the case of Real-Time Multi-Agent Systems the temporal restrictions of their Real-Time Agents make their deliberation process to be temporally bounded. Therefore, this paper presents a guide to temporally bound the CBR to adapt it to be used as deliberative mechanism for Real-Time Agents.

M. Navarro, S. Heras, V. Julián

Towards Providing Social Knowledge by Event Tracing in Multiagent Systems

Social knowledge is one of the key aspects of MAS in order to face complex problems in dynamical environments. However, it is usually incorporated without specific support on behalf of the platform and that does not let agents take all of the advantage of this social knowledge. At present time, the authors of this paper are working in a general tracing system, which could be used by agents in the system to trace other agents’ activity and that could be used as an alternative way for agents to perceive their environment. This paper presents first results of this work, consisting of the requirements which should be taken into account when designing such a tracing system.

Luis Búrdalo, Andrés Terrasa, Ana García-Fornes, Agustín Espinosa

A Solution CBR Agent-Based to Classify SOAP Message within SOA Environments

This paper presents the core component of a solution based on agent technology specifically adapted for the classification of SOA messages. These messages can carry out attacks that target the applications providing Web Services. An advanced mechanism of classification designed in two phases incorporates a CBR-Agent type for classifying the incoming SOAP messages as legal or malicious. Its main feature involves the use of decision trees, fuzzy logic rules and neural networks for filtering attacks.

Cristian Pinzón, Belén Pérez, Angélica González, Ana de Luís y, J. A. Román

RecMas: A Multiagent System Socioconfiguration Recommendations Tool

This paper presents a multiagent recommendation system (RecMAS) able to coordinate the interactions between a user agent (AgUser) and a set of commercial agents (AgComs) providing a useful service for monitoring changes in the AgUser’s beliefs and decisions based on two parameters: (i) the strength of its own beliefs and (ii) the strength of the AgComs’ suggestions. The system was used to test several commercial activities in a shopping centre where the AgComs (AgComs) provided information to an AgUser operating in a wireless device (PDA, mobile phone, etc.) used by a client. The AgUser received messages adapted for conditions of particular offers of interest to the client. Using a theoretical model and a set of simulation experiments, commercial strategies in relation with the socio-dynamics of the system were obtained. This paper concludes with a presentation of a prototype in a real shopping centre.

Luis F. Castillo, Manuel G. Bedia, Ana L. Uribe

Methods of Classifiers Fusion

Combining Multiple Classifiers with Dynamic Weighted Voting

When a multiple classifier system is employed, one of the most popular methods to accomplish the classifier fusion is the simple majority voting. However, when the performance of the ensemble members is not uniform, the efficiency of this type of voting generally results affected negatively. In this paper, new functions for dynamic weighting in classifier fusion are introduced. Experimental results demonstrate the advantages of these novel strategies over the simple voting scheme.

R. M. Valdovinos, J. S. Sánchez

Fusion of Topology Preserving Neural Networks

In this paper ensembles of self organizing NNs through fusion are introduced. In these ensembles not the output signals of the base learners are combined, but their architectures are properly merged. Merging algorithms for fusion and boosting-fusion-based ensembles of SOMs, GSOMs and NG networks are presented and positively evaluated on benchmarks from the UCI database.

C. Saavedra, R. Salas, H. Allende, C. Moraga

Adaptive Splitting and Selection Method of Classifier Ensemble Building

The paper presents a novel machine learning method which allows obtaining compound classifier. Its idea bases on splitting feature space into separate regions and choosing the best classifier from available set of recognizers for each region. Splitting and selection take place simultaneously as a part of an optimization process. Evolutionary algorithm is used to find out the optimal solution. The quality of the proposed method is evaluated via computer experiments.

Konrad Jackowski, Michal Wozniak

Probability Error in Global Optimal Hierarchical Classifier with Intuitionistic Fuzzy Observations

The paper considers the problem of classification error in pattern recognition. This model of classification is primarily based on the Bayes rule and secondarily on the notion of intuitionistic fuzzy sets. A probability of misclassifications is derived for a classifier under the assumption that the features are class-conditionally statistically independent, and we have intuitionistic fuzzy information on object features instead of exact information. Additionally, we consider the global optimal hierarchical classifier.

Robert Burduk

Some Remarks on Chosen Methods of Classifier Fusion Based on Weighted Voting

Multiple Classifier Systems

are nowadays one of the most promising directions in pattern recognition. There are many methods of decision making by the ensemble of classifiers. The most popular are methods that have their origin in voting method, where the decision of the common classifier is a combination of individual classifiers’ outputs. This work presents comparative analysis of some classifier fusion methods based on weighted voting of classifiers’ responses and combination of classifiers’ discriminant functions. We discus which of presented methods could produce classifier better than Oracle one. Some results of computer experiments carried out on benchmark and computer generated data which confirmed our studies are presented also.

Michal Wozniak, Konrad Jackowski

Knowledge Extraction Based on Evolutionary Learning

A Hybrid Bumble Bees Mating Optimization - GRASP Algorithm for Clustering

A new hybrid algorithm for clustering, which is based on the concepts of the Bumble Bees Mating Optimization (BBMO) and Greedy Randomized Adaptive Search Procedure (GRASP), is presented in this paper. The proposed algorithm is a two phase algorithm which combines a new algorithm called Bumble Bees Mating Optimization algorithm for the solution of the feature selection problem and a GRASP algorithm for the solution of the clustering problem. The performance of the algorithm is compared with other popular metaheuristic and nature inspired methods using datasets from the UCI Machine Learning Repository. The high performance of the proposed algorithm is achieved as the algorithm gives very good results and in some instances the percentage of the correct clustered samples is very high and is larger than 98%.

Yannis Marinakis, Magdalene Marinaki, Nikolaos Matsatsinis

A First Study on the Use of Coevolutionary Algorithms for Instance and Feature Selection

Cooperative Coevolution is a technique in the area of Evolutionary Computation. It has been applied to many combinatorial problems with great success. This contribution proposes a Cooperative Coevolution model for simultaneous performing some data reduction processes in classification with nearest neighbours methods through feature and instance selection.

In order to check its performance, we have compared the proposal with other evolutionary approaches for performing data reduction. Results have been analyzed and contrasted by using non-parametric statistical tests, finally showing that the proposed model outperforms the non-cooperative evolutionary techniques.

Joaquín Derrac, Salvador García, Francisco Herrera

Unsupervised Feature Selection in High Dimensional Spaces and Uncertainty

Developing models and methods to manage data vagueness is a current effervescent research field. Some work has been done with supervised problems but unsupervised problems and uncertainty have still not been studied. In this work, an extension of the Fuzzy Mutual Information Feature Selection algorithm for unsupervised problems is outlined. This proposal is a two stage procedure. Firstly, it makes use of the fuzzy mutual information measure and Battiti’s feature selection algorithm and of a genetic algorithm to analyze the relationships between feature subspaces in a high dimensional space. The second stage uses a simple ad hoc heuristic with the aim to extract the most relevant relationships. It is concluded, given the results from the experiments carried out in this preliminary work, that it is possible to apply frequent pattern mining or similar methods in the second stage to reduce the dimensionality of the data set.

José R. Villar, María R. Suárez, Javier Sedano, Felipe Mateos

Non-dominated Multi-objective Evolutionary Algorithm Based on Fuzzy Rules Extraction for Subgroup Discovery

A new multi-objective evolutionary model for subgroup discovery with fuzzy rules is presented in this paper. The method resolves subgroup discovery problems based on the hybridization between fuzzy logic and genetic algorithms, with the aim of extracting interesting, novel and interpretable fuzzy rules. To do so, the algorithm includes different mechanisms for improving diversity in the population. This proposal focuses on the classification of individuals in fronts, based on non-dominated sort. A study can be seen for the proposal and other previous methods for different databases. In this study good results are obtained for subgroup discovery by this new evolutionary model in comparison with existing algorithms.

C. J. Carmona, P. González, M. J. del Jesus, F. Herrera

A First Study on the Use of Interval-Valued Fuzzy Sets with Genetic Tuning for Classification with Imbalanced Data-Sets

Classification with imbalanced data-sets is one of the recent challenging problems in Data Mining. In this framework, the class distribution is not uniform and the separability between the classes is often difficult. From the available techniques in the Machine Learning field, we focus on the use of Fuzzy Rule Based Classification Systems, as they provide an interpretable model for the end user by means of linguistic variables.

The aim of this work is to increase the performance of fuzzy modeling by adding a higher degree of knowledge by means of the use of Interval-valued Fuzzy Sets. Furthermore, we will contextualize the Interval-valued Fuzzy Sets with a post-processing genetic tuning of the amplitude of their upper bounds in order to enhance the global behaviour of this methodology.

J. Sanz, A. Fernández, H. Bustince, F. Herrera

Feature Construction and Feature Selection in Presence of Attribute Interactions

When used for data reduction, feature selection may successfully identify and discard irrelevant attributes, and yet fail to improve learning accuracy because regularities in the concept are still opaque to the learner. In that case, it is necessary to highlight regularities by constructing new characteristics that abstract the relations among attributes. This paper highlights the importance of feature construction when attribute interaction is the main source of learning difficulty and the underlying target concept is hard to discover by a learner using only primitive attributes. An empirical study centered on predictive accuracy shows that feature construction significantly outperforms feature selection because, even when done perfectly, detection of interacting attributes does not sufficiently facilitates discovering the target concept.

Leila S. Shafti, Eduardo Pérez

Multiobjective Evolutionary Clustering Approach to Security Vulnerability Assesments

Network vulnerability assessments collect large amounts of data to be further analyzed by security experts. Data mining and, particularly, unsupervised learning can help experts analyze these data and extract several conclusions. This paper presents a contribution to mine data in this security domain. We have implemented an evolutionary multiobjective approach to cluster data of security assessments. Clusters hold groups of tested devices with similar vulnerabilities to detect hidden patterns. Two different metrics have been selected as objectives to guide the discovery process. The results of this contribution are compared with other single-objective clustering approaches to confirm the value of the obtained clustering structures.

G. Corral, A. Garcia-Piquer, A. Orriols-Puig, A. Fornells, E. Golobardes

Beyond Homemade Artificial Data Sets

One of the most important challenges in supervised learning is how to evaluate the quality of the models evolved by different machine learning techniques. Up to now, we have relied on measures obtained by running the methods on a wide test bed composed of real-world problems. Nevertheless, the unknown inherent characteristics of these problems and the bias of learners may lead to inconclusive results. This paper discusses the need to work under a controlled scenario and bets on artificial data set generation. A list of ingredients and some ideas about how to guide such generation are provided, and promising results of an evolutionary multi-objective approach which incorporates the use of data complexity estimates are presented.

Núria Macià, Albert Orriols-Puig, Ester Bernadó-Mansilla

A Three-Objective Evolutionary Approach to Generate Mamdani Fuzzy Rule-Based Systems

In the last years, several papers have proposed to adopt multi-objective evolutionary algorithms (MOEAs) to generate Mamdani fuzzy rule-based systems with different trade-offs between interpretability and accuracy. Since interpretability is difficult to quantify because of its qualitative nature, several measures have been introduced, but there is no general agreement on any of them. In this paper, we propose an MOEA to learn concurrently rule base and membership function parameters by optimizing accuracy and interpretability, which is measured in terms of number of conditions in the antecedents of rules and partition integrity. Partition integrity is evaluated by using a purposely-defined index based on the piecewise linear transformation exploited to learn membership function parameters. Results on a real-world regression problem are shown and discussed.

Michela Antonelli, Pietro Ducange, Beatrice Lazzerini, Francesco Marcelloni

A New Component Selection Algorithm Based on Metrics and Fuzzy Clustering Analysis

Component-Based Software Engineering is concerned with the assembly of preexisting software components that lead to software systems responding to client specific requirements. This paper presents a new algorithm for constructing a software system by assembling components. The process of selecting a component from a given set takes into account some quality attributes. Metrics are defined in order to quantify the considered attributes. Using these metrics values, a fuzzy clustering approach groups similar components in order to select the best candidate. We comparatively evaluate our results with a case study.

Camelia Şerban, Andreea Vescan, Horia F. Pop

Multi-label Classification with Gene Expression Programming

In this paper, we introduce a Gene Expression Programming algorithm for multi label classification. This algorithm encodes each individual into a discriminant function that shows whether a pattern belongs to a given class or not. The algorithm also applies a niching technique to guarantee that the population includes functions for each existing class. In order to evaluate the quality of our algorithm, its performance is compared to that of four recently published algorithms. The results show that our proposal is the best in terms of accuracy, precision and recall.

J. L. Ávila, E. L. Gibaja, S. Ventura

An Evolutionary Ensemble-Based Method for Rule Extraction with Distributed Data

This paper presents a methodology for knowledge discovery from inherently distributed data without moving it from its original location, completely or partially, to other locations for legal or competition issues. It is based on a novel technique that performs in two stages: first, discovering the knowledge locally and second, merging the distributed knowledge acquired in every location in a common privacy aware maximizing the global accuracy by using evolutionary models. The knowledge obtained in this way improves the one achieved in the local stores, thus it is of interest for the concerned organizations.

Diego M. Escalante, Miguel Angel Rodriguez, Antonio Peregrin

Evolutionary Extraction of Association Rules: A Preliminary Study on their Effectiveness

Data Mining is most commonly used in attempts to induce association rules from transaction data. Most previous studies focused on binary-valued transactions, however the data in real-world applications usually consists of quantitative values. In the last few years, many researchers have proposed Evolutionary Algorithms for mining interesting association rules from quantitative data. In this paper, we present a preliminary study on the evolutionary extraction of quantitative association rules. Experimental results on a real-world dataset show the effectiveness of this approach.

Nicolò Flugy Papè, Jesús Alcalá-Fdez, Andrea Bonarini, Francisco Herrera

A Minimum-Risk Genetic Fuzzy Classifier Based on Low Quality Data

Minimum risk classification problems use a matrix of weights for defining the cost of misclassifying an object. In this paper we extend a simple genetic fuzzy system (GFS) to this case. In addition, our method is able to learn minimum risk fuzzy rules from low quality data. We include a comprehensive description of the new algorithm and discuss some issues about its fuzzy-valued fitness function. A synthetic problem, plus two real-world datasets, are used to evaluate our proposal.

Ana M. Palacios, Luciano Sánchez, Inés Couso

Hybrid Systems Based on Bioinspired Algorithms and Argumentation Models

Performance Analysis of the Neighboring-Ant Search Algorithm through Design of Experiment

In many science fields such as physics, chemistry and engineering, the theory and experimentation complement and challenge each other. Algorithms are the most common form of problem solving in many science fields. All algorithms include parameters that need to be tuned with the objective of optimizing its processes. The NAS (Neighboring-Ant Search) algorithm was developed to route queries through the Internet. NAS is based on the ACS (Ant Colony System) metaheuristic and SemAnt algorithm, hybridized with local strategies such as: learning, characterization, and exploration. This work applies techniques of

Design of Experiments

for the analysis of NAS algorithm. The objective is to find out significant parameters for the algorithm performance and relations among them. Our results show that the probability distribution of the network topology has a huge significance in the performance of the NAS algorithm. Besides, the probability distributions of queries invocation and repositories localization have a combined influence in the performance.

Claudia Gómez Santillán, Laura Cruz Reyes, Eustorgio Meza Conde, Claudia Amaro Martinez, Marco Antonio Aguirre Lam, Carlos Alberto Ochoa Ortíz Zezzatti

A New Approach to Improve the Ant Colony System Performance: Learning Levels

In this paper a hybrid ant colony system algorithm is presented. A new approach to update the pheromone trails, denominated learning levels, is incorporated. Learning levels is based on the distributed Q-learning algorithm, a variant of reinforcement learning, which is incorporated to the basic ant colony algorithm. The hybrid algorithm is used to solve the Vehicle Routing Problem with Time Windows. Experimental results with the Solomon’s dataset of instances reveal that learning levels improve execution time and quality, respect to the basic ant colony system algorithm, 0.15% for traveled distance and 0.6% in vehicles used. Now we are applying the hybrid ant colony system in other domains.

Laura Cruz R., Juan J. Gonzalez B., José F. Delgado Orta, Barbara A. Arrañaga C., Hector J. Fraire H.

Hybrid Algorithm to Data Clustering

In this research an N-Dimentional clustering algorithm based on ACE algorithm for large datasets is described. Each part of the algorithm will be explained and experimental results obtained from apply this algorithm are discussed. The research is focused on the fast and accurate clustering using real databases as workspace instead of directly loaded data into memory since this is very limited and insufficient when large data amount are used. This algorithm can be applied to a great variety and types of information i.e. geospatial data, medical data, biological data and others. The number of computations required by the algorithm is ~O(N).

Miguel Gil, Alberto Ochoa, Antonio Zamarrón, Juan Carpio

Hybrid Evolutionary Intelligence in Financial Engineering

Financial Forecasting of Invoicing and Cash Inflow Processes for Fair Exhibitions

The concept of case-based reasoning (CBR) system for financial forecasting, the dynamics of issuing invoices and the corresponding cash inflow for a fair exhibition is presented in this paper. The aim of the research is the development of a hybrid intelligent tool, based on the heuristic interpolating CBR


phase and the data gravitation classification method in the


phase. The previous experience with new forecasting is taken into account. Simulations performed are based on the already known behaviour of the adapted logistic curves followed during the last eight years of fair exhibitions. Methodological aspects have been practically tested as a part of the management information system development project of ”Novi Sad Fair”.

Dragan Simić, Ilija Tanackov, Vladeta Gajić, Svetlana Simić

A Hybrid Neural Network-Based Trading System

We present a hybrid intelligent trading system that combines artificial neural networks (ANN) and particle swarm optimisation (PSO) to generate optimal trading decisions. A PSO algorithm is used to train ANNs using objective functions that are directly linked to the performance of the trading strategy rather than statistical measures of forecast error (e.g. mean squared error). We experiment with several objective measures that quantify the return/risk associated with the trading system. First results from the application of this methodology to real data show that the out-of-sample performance of trading models is fairly consistent with respect to the objective function they derive from.

Nikos S. Thomaidis, Georgios D. Dounias

Active Portfolio Management under a Downside Risk Framework: Comparison of a Hybrid Nature – Inspired Scheme

Hybrid intelligent systems are becoming more and more popular in solving nondeterministic polynomial-time – hard optimization problems. Lately, the focus is on nature – inspired intelligent algorithms, whose main advantage is the exploitation of unique features of natural systems. One type of complex optimization problems is the active portfolio management, where the incorporation of complex, realistic constraints makes it difficult for traditional numerical methods to deal with it. In this paper we perform a computational study of a hybrid Ant Colony Optimization algorithm. The application is a specific formulation of the problem. Our main aim in this paper is to introduce a new framework of study in the field of active portfolio management, where the main interest lies in minimizing the risk of the portfolio return falling below the benchmark. Secondary, we provide some preliminary results regarding the use of a new hybrid nature – inspired scheme in solving this type of problem.

Vassilios Vassiliadis, Nikolaos Thomaidis, George Dounias


Additional information

Premium Partner

    Image Credits