Artificial Intelligence Applications and Innovations

In this paper we present a new method for robot real time policy adaptation by combining learning and evolution. The robot adapts the policy as the environment conditions change. In our method, we apply evolutionary computation to find the optimal relation between reinforcement learning parameters and robot performance. The proposed algorithm is evaluated in the simulated environment of the Cyber Rodent (CR) robot, where the robot has to increase its energy level by capturing the active battery packs. The CR robot lives in two environments with different settings that replace each other four times. Results show that evolution can generate an optimal relation between the robot performance and exploration-exploitation of reinforcement learning, enabling the robot to adapt online its strategy as the environment conditions change.

Theories of embodied cognition and active vision suggest that perception is constructed through interaction and becomes meaningful because it is grounded in the agent’s activity. We developed a model to illustrate and implement these views. Following its intrinsic motivation, the agent autonomously learns to coordinate its motor actions with the information received from its sensory system. Besides illustrating theories of active vision, this model suggests new ways to implement vision and intrinsic motivation in artificial systems. Specifically, we coupled an intrinsically motivated schema mechanism with a visual system. To connect vision with sequences, we made the visual system react to movements in the visual field rather than merely transmitting static patterns.

This paper reports an extension of the previous MIT and Caltech’s cortex-like machine vision models of Graph-Based Visual Saliency (GBVS) and Feature Hierarchy Library (FHLIB), to remedy some of the undesirable drawbacks in these early models which improve object recognition efficiency. Enhancements in three areas, a) extraction of features from the most salient region of interest (ROI) and their rearrangement in a ranked manner, rather than random extraction over the whole image as in the previous models, b) exploitation of larger patches in the C1 and S2 layers to improve spatial resolutions, c) a more versatile template matching mechanism without the need of ‘pre-storing’ physical locations of features as in previous models, have been the main contributions of the present work. The improved model is validated using 3 different types of datasets which shows an average of ~7% better recognition accuracy over the original FHLIB model.

A novel evaluation metric is introduced, based on the Discernibility concept. This metric, the Distance-based Index of Discernibility (DID) aims to provide an accurate and fast mapping of the classification performance of a feature or a dataset. DID has been successfully implemented in a program which has been applied to a number of datasets, a few artificial features and a typical benchmark dataset. The results appear to be quite promising, verifying the initial hypothesis.

This article introduces the concept of the variability of association rules of products through the estimate of a new indicator called overall variability of association rules (OCVR). The proposed indicator applied to super market chain products, tries to highlight product market baskets, with great variability in consumer behavior. Parameter of the variability of association rules in connection with changes in the purchasing habit during the course of time, can contribute further to the efficient market basket analysis and appropriate marketing strategies to promote sales. These strategies may include changing the location of the products on the shelf, the redefinition of the discount or even policy or even the successful of recommendation systems.

This paper proposes the Java Evolutionary Computation Library (JECoLi), an adaptable, flexible, extensible and reliable software framework implementing metaheuristic optimization algorithms, using the Java programming language. JECoLi aims to offer a solution suited for the integration of Evolutionary Computation (EC)-based approaches in larger applications, and for the rapid and efficient benchmarking of EC algorithms in specific problems. Its main contributions are (i) the implementation of pluggable parallelization modules, independent from the EC algorithms, allowing the programs to adapt to the available hardware resources in a transparent way, without changing the base code; (ii) a flexible platform for software quality assurance that allows creating tests for the implemented features and for user-defined extensions. The library is freely available as an open-source project.

London Plan is the London mayor’s Spatial Development Strategy. This strategic long-term plan comprises of proposals for different aspects of change within the London boundary. Furthermore, the proposals include chapters outlining adjustments in each facet. Policy 4A.2 reflects the Climate Change Mitigation scheme. Some consultations and research works have been performed to this point, but an extensive cost assessment has not been done. This paper reflects a financial assessment by means of Influence Diagrams based upon the London Plan policies 4A.X.

To ensure long-term competitiveness, companies try to maintain a high level of agility, flexibility and responsiveness. In many domains, hierarchical SCs are considered as dynamic systems that deal with many perturbations. In this paper, we handle a specific type of supply chain: a Crisis Management Supply Chain (CMSC). Supply during peacetime can be managed by proactive logistics plans and classic supply chain management techniques to guaranty the availability of required needs. However, in case of perturbations (time of war, natural disasters…) the need for support increases dramatically and logistics plans need to be adjusted rapidly. Subjective variables like risk, uncertainty and vulnerability will be used in conjunction with objective variables such as inventory levels, delivery times and financial loss to determine preferred courses of action.

Contemporary Project and Portfolio Management Information Systems (PPMIS) have embarked from single-user, single-project management systems to web-based, collaborative, multi-project, multi-functional information systems which offer organization-wide management support. The variety of offered functionalities along with the variation among each organization needs and the plethora of PPMIS available in the market, make the selection of a proper PPMIS a difficult, multi-criteria decision problem. The problem complexity is further augmented since the multi stakeholders involved cannot often rate precisely their preferences and the performance of candidate PPMIS on them. To meet these challenges, this paper presents a PPMIS selection/evaluation approach that combines TOPSIS (Technique for Order Preference by Similarity to Ideal Solution) with intuitionistic fuzzy group decision making. The approach considers the vagueness of evaluators’ assessments when comparing PPMIS and the uncertainty of users to judge their needs.

In this paper, we present the design, implementation and evaluation of intelligent methods that assess bank loan applications. Assessment concerns the ability/possibility of satisfactorily dealing with loan demands. Different loan programs from different banks may be proposed according to the applicant’s characteristics. For each loan program, corresponding attributes (e.g. interest, amount of money that can be loaned) are also calculated. For these tasks, two separate intelligent systems have been developed and evaluated: a fuzzy expert system and a neuro-symbolic expert system. The former employs fuzzy rules based on knowledge elicited from experts. The latter is based on neurules, a type of neuro-symbolic rules that combine a symbolic (production rules) and a connectionist (adaline unit) representation. Neurules were produced from available patterns. Evaluation showed that performance of both systems is close although their knowledge bases were derived from different types of source knowledge.

The paper presents IF-inference systems of Takagi-Sugeno type. It is based on intuitionistic fuzzy sets (IF-sets), introduced by K.T. Atanassov, fuzzy t-norm and t-conorm, intuitionistic fuzzy t-norm and t-conorm. Thus, an IF-inference system is developed for ozone time series prediction. Finally, we compare the results of the IF-inference systems across various operators.

Analysis of attitude stabilization of a power-aided unicycle points out that a unicycle behaves like an inverted pendulum subject to power constraint. An LQR-mapped fuzzy controller is introduced to solve this nonlinear issue by mapping LQR control reversely through least square and Sugeno-type fuzzy inference. The fuzzy rule surface after mapping remains optimal.

Dealing with a LQR controller surface subject to power and torque constraints, is an issue of nonlinear problem that is difficult to implement. This paper employs a fuzzy controller surface to replace the LQR surface subject to power and torque constraints by using class stacking, least square and Sugeno-type fuzzy inference mode. Through this type of transformation, called “Optimal fuzzy controller mapped from LQR”, control of the system remains optimal.

In this paper a new criterion for clusters validation is proposed. This new cluster validation criterion is used to approximate the goodness of a cluster. The clusters which satisfy a threshold of this measure are selected to participate in clustering ensemble. For combining the chosen clusters, a co-association based consensus function is applied. Since the Evidence Accumulation Clustering method cannot derive the co-association matrix from a subset of clusters, a new EAC based method which is called Extended EAC, EEAC, is applied for constructing the co-association matrix from the subset of clusters. Employing this new cluster validation criterion, the obtained ensemble is evaluated on some well-known and standard data sets. The empirical studies show promising results for the ensemble obtained using the proposed criterion comparing with the ensemble obtained using the standard clusters validation criterion.

Dynamical modeling is used to describe the process of animals’ repeated learning. Theoretical analysis is done to explore the dynamic property of this process, such as the limit sets and their stability. The scope of variables is provided for different practical purpose, cooperated with necessary numerical simulation.

In the last decades, a myriad of approaches to the multi-armed bandit problem have appeared in several different fields. The current top performing algorithms from the field of Learning Automata reside in the Pursuit family, while UCB-Tuned and the

ε

-greedy class of algorithms can be seen as state-of-the-art regret minimizing algorithms. Recently, however, the Bayesian Learning Automaton (BLA) outperformed all of these, and other schemes, in a wide range of experiments. Although seemingly incompatible, in this paper we integrate the foundational learning principles motivating the design of the BLA, with the principles of the so-called Generalized Pursuit algorithm (GPST), leading to the

Generalized Bayesian Pursuit

algorithm (GBPST). As in the BLA, the estimates are truly Bayesian in nature, however, instead of basing exploration upon direct sampling from the estimates, GBPST explores by means of the arm selection probability vector of GPST. Further, as in the GPST, in the interest of higher rates of learning, a

set of arms

that are currently perceived as being optimal is pursued to minimize the probability of pursuing a wrong arm. It turns out that GBPST is superior to GPST and that it even performs better than the BLA by controlling the learning speed of GBPST. We thus believe that GBPST constitutes a new avenue of research, in which the performance benefits of the GPST and the BLA are mutually augmented, opening up for improved performance in a number of applications, currently being tested.

The Quadratic Assignment Problem (QAP) is an NP-complete problem. Different algorithms have been proposed using different methods. In this paper, the problem is formulated as a minimizing problem of a quadratic function with restrictions incorporated to the computational dynamics and variables Si ∈{1,2,..., n}. To solve this problem a recurrent neural network multivalued (RNNM) is proposed. We present four computational dynamics and we demonstrate that the energy of the neuron network decreases or remains constant according to the Computer Dynamic defined.

This paper aims in comparing countries with different energy strategies, and demonstrate the close connection between environment and economic growth in the ex-Eastern countries, during their transition to market economies. We have developed a radial-basis function neural network system, which is trained to classify countries based on their emissions of carbon, sulphur and nitrogen oxides, and on their Gross National Income. We used three countries representative of ex-Eastern economies (Russia, Poland and Hungary) and three countries representative of Western economies (United States, France and United Kingdom). Results showed that the linkage between environmental pollution and economic growth has been maintained in ex-Eastern countries.

Declarative Modelling environments exhibit an idiosyncrasy that demands specialised machine learning methodologies. The particular characteristics of the datasets, their irregularity in terms of class representation, volume, availability as well as user induced inconsistency further impede the learning potential of any employed mechanism, thus leading to the need for adaptation and adoption of custom approaches, expected to address these issues. In the current work we present the problems encountered in the effort to acquire and apply user profiles in such an environment, the modified boosting learning algorithm adopted and the corresponding experimental results.

The ability to timely predict the academic performance tendency of postgraduate students is very important in MSc programs and useful for tutors. The scope of this research is to investigate which is the most efficient machine learning technique in predicting the final grade of Ionian University Informatics postgraduate students. Consequently, five academic courses are chosen, each constituting an individual dataset, and six well-known classification algorithms are experimented with. Furthermore, the datasets are enriched with demographic, in-term performance and in-class behaviour features. The small size of the datasets and the imbalance in the distribution of class values are the main research challenges of the present work. Several techniques, like resampling and feature selection, are employed to address these issues, for the first time in a performance prediction application. Naïve Bayes and 1-NN achieved the best prediction results, which are very satisfactory compared to those of similar approaches.

Software development organisations are under heavy pressure to complete projects on time, within budget and with the appropriate level of quality, and many questions are asked when a project fails to meet any or all of these requirements. Over the years, much research effort has been spent to find ways to mitigate these failures, the reasons of which come from both within and outside the organisation’s control. One possible risk of failure lies in human resource management and, since humans are the main asset of software organisations, getting the right team to do the job is critical. This paper proposes a procedure for software project managers to support their project scheduling and team staffing activities – two areas where human resources directly impact software development projects and management decisions – by adopting a genetic algorithm approach as an optimisation technique to help solve software project scheduling and team staffing problems.

Similarity measurement between two musical pieces is a hard problem. Humans perceive such similarity by employing a large amount of contextually semantic information. Commonly used content-based me-thodologies rely on information that includes little or no semantic information, and thus are reaching a performance “upper bound”. Recent research pertaining to contextual information assigned as free-form text (tags) in social networking services has indicated tags to be highly effective in improving the accuracy of music similarity. In this paper, we perform a large scale (20k real music data) similarity measurement using mainstream content and context methodologies. In addition, we test the accuracy of the examined methodologies against not only objective metadata but real-life user listening data as well. Experimental results illustrate the conditionally substantial gains of the context-based methodologies and a not so close match these methods with the real user listening data similarity.

The present work aims to create a shallow parser for Modern Greek subject/object detection, using machine learning techniques. The parser relies on limited resources. Experiments with equivalent input and the same learning techniques were conducted for English, as well, proving that the methodology can be adjusted to deal with other languages with only minor modifications. For the first time, the class imbalance problem concerning Modern Greek syntactically annotated data is successfully addressed.

Greeklish to Greek transcription does undeniably seem to be a challenging task since it cannot be accomplished by directly mapping each Greek character to a corresponding symbol of the Latin alphabet. The ambiguity in the human way of Greeklish writing, since Greeklish users do not follow a standardized way of transliteration makes the process of transcribing Greeklish back to Greek alphabet challenging. Even though a plethora of deterministic approaches for the task at hand exists, this paper presents a non-deterministic, vocabulary-free approach, which produces comparable and even better results, supports argot and other linguistic peculiarities, based on an ensemble classification methodology of Data Mining, namely Random Forests. Using data from real users from a conglomeration of resources such as Blogs, forums, email lists, etc., as well as artificial data from a robust stochastic Greek to Greeklish transcriber, the proposed approach depicts satisfactory outcomes in the range of 91.5%-98.5%, which is comparable to an alternative commercial approach.

In this work we formalize a natural expansion of timed argumentation frameworks by considering arguments that are available with (possibly) some repeated interruptions in time, called

intermittent arguments

. This framework is used as a modelization of argumentation dynamics. The notion of acceptability of arguments is analyzed as the framework evolves through time, and an algorithm for computing intervals of argument defense is introduced.

Specialized applied fields in natural sciences – medicine, biology, chemistry etc. require building and exploring of information systems based on related text forms (words, phrases). These forms represent expert information and knowledge. The paper discusses the integration of two basic approaches – relational for structuring complex related texts and object oriented for data analysis. This conception is implemented for building of information system “Crop protection” in Bulgaria based on the complex relationships between biological (crops, pests) and chemical (pesticides) terms in textual form. Analogy exists between class objects in biology, chemistry and class objects and instances of object oriented programming. That fact is essential for building flexible models and software for data analysis in the information system. The presented example shows the potential of object oriented modelling to define and resolve complex tasks concerning effective pesticides use.

Large state spaces pose a serious problem in many learning applications. This paper discusses a number of issues that arise when ranking functions are applied to such a domain. Since these functions, in their original introduction, need to store every possible world model, it seems obvious that they are applicable to small toy problems only. To disprove this we address a number of these issues and furthermore describe an application that indeed has a large state space. It is shown that an agent is

enabled

to learn in this environment by representing its belief state with a ranking function. This is achieved by introducing a new entailment operator that accounts for similarities in the state description.

The paper presents basic notions of web mining, radial basis function (RBF) neural networks and

ε

-insensitive support vector machine regression (

ε

- SVR) for the prediction of a time series for the website of the University of Pardubice. The model includes pre-processing time series, design RBF neural networks and

ε

-SVR structures, comparison of the results and time series prediction. The predictions concerning short, intermediate and long time series for various ratios of training and testing data. Prediction of web data can be benefit for a web server traffic as a complicated complex system.

We propose a framework for predicting the ranking position of a Web page based on previous rankings. Assuming a set of successive top-k rankings, we learn predictors based on different methodologies.

The prediction quality is quantified as the similarity between the predicted and the actual rankings. Extensive experiments were performed on real world large scale datasets for global and query-based top-k rankings, using a variety of existing similarity measures for comparing top-k ranked lists, including a novel and more strict measure introduced in this paper. The predictions are highly accurate and robust for all experimental setups and similarity measures.

The calibration method used in this study allows for the examination of distributed, but potentially subtle, representations of semantic information between mechanistic encoding of the language and the EEG. In particular, a horizontal connection between two basic Fundamental Operations (Semantic Composition and Synchronization) is attempted. The experimental results gave significant differences, which can be considered reliable and promising for further investigation. The experiments gave helpful results. Consequently, this method will be tested along with the classification step by appropriate neural network classifiers.

Parkinson’s disease is associated with motor symptoms, including tremor. The DBS (Deep Brain Stimulation) involves electrode implantation into sub-cortical structures for long-term stimulation at frequencies greater than 100Hz. We performed linear and nonlinear analysis of the tremor signals to determine a set of parameters and rules for recognizing the behavior of the investigated patient and to characterize the typical responses for several forms of DBS. We found patterns for homogeneous group for data reduction. We used Data Mining and Knowledge discovery techniques to reduce the number of data. To support such predictions, we develop a model of the tremor, to perform tests determining the DBS reducing the tremor or inducing tolerance and lesion if the stimulation is chronic.

A major drawback of most existing medical decision support systems is that they do not provide any indication about the uncertainty of each of their predictions. This paper addresses this problem with the use of a new machine learning framework for producing valid probabilistic predictions, called Venn Prediction (VP). More specifically, VP is combined with Neural Networks (NNs), which is one of the most widely used machine learning algorithms. The obtained experimental results on two medical datasets demonstrate empirically the validity of the VP outputs and their superiority over the outputs of the original NN classifier in terms of reliability.

Window Memoization

is a performance improvement technique for image processing algorithms. It is based on removing computational redundancy in an algorithm applied to a single image, which is inherited from data redundancy in the image. The technique employs a fuzzy reuse mechanism to eliminate unnecessary computations. This paper extends the window memoization technique such that in addition to exploiting the data redundancy in a single image, the data redundancy in a sequence of images of a volume data is also exploited. The detection of the additional data redundancy leads to higher speedups. The cascaded window memoization technique was applied to Canny edge detection algorithm where the volume data of prostate MR images were used. The typical speedup factor achieved by cascaded window memoization is 4.35x which is 0.93x higher than that of window memoization.

In this work, we investigate a fast background elimination front-end of an automatic bacilli detection system. This background eliminating system consists of a feature descriptor followed by a linear-SVMs classifier. Four state-of-the-art feature extraction algorithms are analyzed and modified. Extensive experiments have been made on real sputum fluorescence images and the results reveal that 96.92% of the background content can be correctly removed from one image with an acceptable computational complexity.

The article presents assumptions, and results of a test of experimental verification of a hypothesis stating that the use of the MammoEdit - a tool that uses its own ontology and the embedded knowledge of mammography – increase the diagnostic accuracy as well as the reproducibility of mammographic interpretation. The graphical user interface of the editor was similarly assessed, as well as the rules for visualization which assists the radiologist in the interpretation of the lesions’ character.

While the plethora of artificial biomedical applications is enriched and combined with the possibilities of artificial intelligence, bioinformatics and nanotechnology, the variability in the ideological use of such concepts is associated with bioethical issues and several legal aspects. The convergence of bioethics and computer ethics, attempts to illustrate and approach problems, occurring by the fusion of human and machine or even through the replacement of human determination by super intelligence. Several issues concerning the effects of artificial biomedical applications will be discussed, considering the upcoming post humanism period.

Recycling has been gaining ground, thanks to the recent progress made in the related technology. However, a limiting factor to its wide adoption, is the lack of modern tools for managing the collection of recyclable resources. In this paper, we present EcoTruck, a management system for the collection of recyclable paper products. EcoTruck is modelled as a multi-agent system and its implementation employs Erlang, a distribution-oriented declarative language. The system aims to automate communication and cooperation of parties involved in the collection process, as well as optimise vehicle routing. The latter have the effect of minimising vehicle travel distances and subsequently lowering transportation costs. By speeding up the overall recycling process, the system could increase the service throughput, eventually introducing recycling methods to a larger audience.

Artificial Intelligence has been present since more than two decades ago, in the treatment of data concerning the protection of the environment; in particular, various groups of researchers have used genetic algorithms and artificial neural networks in the analysis of data related to the atmospheric sciences and the environment. However, in this kind of applications has been conspicuously absent from the associative models, by virtue of which the classic associative techniques exhibit very low yields. This article presents the results of applying Alpha-Beta associative models in the analysis and prediction of the levels of Carbon Monoxide (CO) and Nitrogen Oxides (NO

x

) in Mexico City.

The paper presents a method for automatic detection and monitoring of small waterlogged areas in farmland, using multispectral satellite images and neural network classifiers. In the waterlogged areas, excess water significantly damages or completely destroys the plants, thus reducing the average crop yield. Automatic detection of (waterlogged) crops damaged by rising underground water is an important tool for government agencies dealing with yield assessment and disaster control.

The paper describes the application of two different neural network algorithms to the problem of identifying crops that have been affected by rising underground water levels in WorldView-2 satellite imagery. A satellite image of central European region (North Serbia), taken in May 2010, with spatial resolution of 0.5

m

and 8 spectral bands was used to train the classifiers and test their performance when it comes to identifying the water-stressed crops. WorldView-2 provides 4 new bands potentially useful in agricultural applications: coastal-blue, red-edge, yellow and near-infrared 2. The results presented show that a Multilayer Perceptron is able to identify the damaged crops with 99.4% accuracy. Surpassing previously published methods.

This paper aims in the reduction of data attributes of a fuzzy-set based system for the estimation of forest fire risk in Greece, with the use of rough-set theory. The aim is to get as good results as possible with the use of the minimum amount of data attributes possible. Data manipulation for this project is done in MS-Access. The resulting data table is inserted into Matlab in order to be fuzzified. The final result of this clustering is inserted into Rossetta, which is a Rough set exploration software, in order to estimate the reducts. The risk estimation is recalculated with the use of the reduct set in order to measure the accuracy of the final minimum attribute set. Nine forest fire risk factors were taken into consideration for the purpose of this paper and the Greek terrain was separated into smaller areas, each concerning a different Greek forest department.

Saving earth’s biodiversity for future generations is an important global task, where automatic recognition of pollen species by means of computer vision represents a highly prioritized issue. This work focuses on analysis and classification stages. A combination of geometrical measures, Fourier descriptors of morphological details using Discrete Cosine Transform (DCT) in order to select their most significant values, and colour information over decorrelated stretched images are proposed as pollen grains discriminative features. A Multi-Layer neural network was used as classifier applying scores fusion techniques. 17 tropical honey plant species have been classified achieving a mean of 96.49% ± 1.16 of success.

Estimation by Analogy is a popular method in the field of software cost estimation. A number of research approaches focus on optimizing the parameters of the method. This paper proposes an optimal global setup for determining empirically the best method parameter configuration based on genetic algorithms. We describe how such search can be performed, and in particular how spaces whose dimensions are of different type can be explored. We report results on two datasets and compare with approaches that explore partially the search space. Results provide evidence that our method produces similar or better accuracy figures with respect to other approaches.

Software project management makes extensive use of predictive modeling to estimate product size, defect proneness and development effort. Although uncertainty is acknowledged in these tasks, fuzzy inference systems, designed to cope well with uncertainty, have received only limited attention in the software engineering domain. In this study we empirically investigate the impact of two choices on the predictive accuracy of generated fuzzy inference systems when applied to a software engineering data set: sampling of observations for training and testing; and the size of the rule set generated using fuzzy c-means clustering. Over ten samples we found no consistent pattern of predictive performance given certain rule set size. We did find, however, that a rule set compiled from multiple samples generally resulted in more accurate predictions than single sample rule sets. More generally, the results provide further evidence of the sensitivity of empirical analysis outcomes to specific model-building decisions.

With ever increasing application of information technologies in every day activities, organizations face the need for applications that provides better security. The existence of complex IT systems with multiple interdependencies creates great difficulties for Chief Security Officers to comprehend and be aware of all potential risks in such systems. Intelligent decision making for IT security is a crucial element of an organization’s success and its competitive position in the marketplace. This paper considers the implementation of an integrated attack graph and a Fuzzy Cognitive Maps (FCM) to provide facilities to capture and represent complex relationships in IT systems. By using FCMs the security of IT systems can regularly be reviewed and improved. What-if analysis can be performed to better understand vulnerabilities of a designed system. Finally an integrated system consisting of FCM, Attack graphs and Genetic Algorithms (GA) is used to identify vulnerabilities of IT systems that may not be apparent to Chief Security Officers.

In the field of search based software engineering, evolutionary testing is a very popular domain in which test cases are automatically generated for a given piece of code using evolutionary algorithms. The techniques used in this domain usually are hard to compare since there is no standard testbed. In this paper we propose an automatic program generator to solve this situation. The program generator is able to create Java programs with the desired features. In addition, we can ensure that all the branches in the programs are reachable, i.e. a 100% branch coverage is always possible. Thanks to this feature the research community can test and enhance their algorithms until a total coverage is achieved. The potential of the program generator is illustrated with an experimental study on a benchmark of 800 generated programs. We highlight the correlations between some static measures computed on the program and the code coverage when an evolutionary test case generator is used. In particular, we compare three techniques as the search engine for the test case generator: an Evolutionary Strategy, a Genetic Algorithm and a Random Search.

When building a data pipeline to process medical claims there are many instances where automated classification schemes could be used to improve speed and efficiency. Medical bills can be classified by the statutory environment which determines appropriate adjudication of payment disputes. We refer to this classification result as the

adjudication type

of a bill. This classification can be used to determine appropriate payment for medical services.

Using a set of 182,811 medical bills, we develop a procedure to quickly and accurately determine the correct adjudication type. A simple naïve Bayes classifier based on training set class occurrences gives 92.8% accuracy, which can be remarkably improved by instead presenting these probabilities to an artificial neural network, yielding 96.8 ±0.5 % accuracy.

This study investigates the application of classification methods for the prognosis of future disability on MRI-detectable brain white matter lesions in subjects diagnosed with clinical isolated syndrome (CIS) of multiple sclerosis (MS). For this purpose, MS lesions and normal appearing white matter (NAWM) from 30 symptomatic untreated MS subjects, as well as normal white matter (NWM) from 20 healthy volunteers, were manually segmented, by an experienced MS neurologist, on transverse T2-weighted images obtained from serial brain MR imaging scans. A support vector machines classifier (SVM) based on texture features was developed to classify MRI lesions detected at the onset of the disease into two classes, those belonging to patients with EDSS≤2 and EDSS>2 (expanded disability status scale (EDSS) that was measured at 24 months after the onset of the disease). The highest percentage of correct classification’s score achieved was 77%. The findings of this study provide evidence that texture features of MRI-detectable brain white matter lesions may have an additional potential role in the clinical evaluation of MRI images in MS. However, a larger scale study is needed to establish the application of texture analysis in clinical practice.

This paper aims to discuss engineering aspects for an agent-based ambient assisted living system for the home environment using argumentation for decision making. The special requirements of our system are to provide a platform with cost-effective specialized assisted living services for the elderly people having cognitive problems, which will significantly improve the quality of their home life, extend its duration and at the same time reinforce social networking. The proposed architecture is based on an agent platform with personal assistant agents that can service users with more than one type of health problems.

Information on the applied pressure is critical to the pressure garment treatment. The use of the passive resonance sensors would be significant improvement to existing systems. These sensors have nonlinear response and thus require nonlinear regression methods. In this paper we compare three nonlinear modelling methods: Sugeno type fuzzy inference system, support vector regression and multilayer perception networks. According to the results, all the tested methods are adequate for modelling an individual sensor. The used methods also give promising results when they are used to model responses of multiple sensors.

Sensor data generated by pervasive applications are very diverse and are rarely described in standard or established formats. Consequently, one of the greatest challenges in pervasive systems is to integrate heterogeneous repositories of sensor data into a single view. The traditional approach to data integration, where a global schema is designed to incorporate the local schemas, may not be suitable to sensor data due to their highly transient schemas and formats. Furthermore, researchers and professionals in healthcare need to combine relevant data from various data streams and other data sources, and to be able to perform searches over all of these collectively using a single interface or query. Often, users express their search in terms of a small set of predefined fields from a single schema that is the most familiar to them, but they want their search results to include data from other compatible schemas as well. We have designed and implemented a framework for a sensor data repository that gives access to heterogeneous sensor metadata schemas in a uniform way. In our framework, the user specifies a query in an arbitrary schema and specifies the mappings from this schema to all the collections he wants to access. To ease the task of mapping specification, our system remembers metadata mappings previously used and uses them to propose other relevant mapping choices for the unmapped metadata elements. That way, users may build their own metadata mappings based on earlier mappings, each time specifying (or improving) only those components that are different. We have created a repository using data collected from various pervasive applications in a healthcare environment, such as activity monitoring, fall detection, sleep-pattern identification, and medication reminder systems, which are currently undergoing at the Heracleia Lab. We have also developed a flexible query interface to retrieve relevant records from the repository that allows users to specify their own choices of mappings and to express conditions to effectively access fine-grained data.

In this work we consider the problem of feature selection in the context of conformal prediction. Unlike many conventional machine learning methods, conformal prediction allows to supply individual predictions with valid measure of confidence. The main idea is to use confidence measures as an indicator of usefulness of different features: we check how many features are enough to reach desirable average level of confidence. The method has been applied to abdominal pain data set. The results are discussed.

Conformal prediction is a recently developed flexible method which allows making valid predictions based on almost any underlying classification or regression algorithm. In this paper, conformal prediction technique is applied to the problem of diagnosing Bovine Tuberculosis. Specifically, we apply Nearest-Neighbours Conformal Predictor to the VETNET database in an attempt to allow the increase of the positive prediction rate of the existing Skin Test. Conformal prediction framework allows us to do so while controlling the risk of misclassifying true positives.

We propose a methodological framework for the classification of the tree-like structures of the ductal network of human breast regarding radiological findings related to breast cancer. Initially we perform the necessary preprocessing steps such as image segmentation in order to isolate the ductal tree structure from the background of x-ray galactograms. Afterwards, we employ tree characterization approaches to obtain a symbolic representation of the distribution of trees’ branching points. Our methodology is based on Sholl analysis, a technique which uses concentric circles that radiate from the center of the region of interest. Finally, we apply the k-nearest neighbor classification scheme to characterize the tree-like ductal structures in galactograms in order to distinguish among different radiological findings. The experimental results are quite promising as the classification accuracy reaches up to 82% indicating that our methods may assist radiologists to identify image biomarkers in galactograms.

Obstructive Nephropathy (ON) is a renal disease and its pathology is believed to be magnified by various molecular processes. In the current study, we apply an intelligent workflow implemented in Rapidminer data mining platform to two different ON datasets. Our scope is to select the most important actors in two corresponding molecular information levels: human miRNA and mouse mRNA. A forward selection method with an embedded nearest neighbor classifier is initially applied to select the most important features in each level. The resulting features are next fed to classifiers appropriately tested utilizing a leave-one-out resampling technique in order to evaluate the relevance of the selected input features when used to classify subjects into output classes defined by ON severity. Preliminary results show that high classification accuracies are obtained, and are supported by the fact that the selected miRNAs or mRNAs have been found significant within differential expression analysis using the same datasets.

In this paper, we propose a clustering technique for the recognition of pigmented skin lesions in dermatological images. It is known that computer vision-based diagnosis systems have been used aiming mostly at the early detection of skin cancer and more specifically the recognition of malignant melanoma tumor. The feature extraction is performed utilizing digital image processing methods, i.e. segmentation, border detection, color and texture processing. The proposed method combines an already successful clustering technique from the field of projection based clustering with a projection pursuit method. Experimental results show great performance on detecting the skin cancer.

The main aim of this paper is to compare the results of several methods of prediction with confidence. In particular we compare the results of Venn Machine with Platt’s Method of estimating confidence. The results are presented and discussed.