Modelling and Implementation of Complex Systems

Cloud Computing provides on demand resources for customers and enterprises to outsource their online activities efficiently and less expensively. However, the cloud environment is heterogeneous and very dynamic, storage node failures and increasing demands on data can lead to data unavailability situations leading to a decrease in quality of service. Cloud service providers face the challenge of ensuring maximum data availability and reliability. Replication of data to different nodes in the cloud has become the most common solution for achieving good performance in terms of load balancing, response time and availability. In this article, we propose a new dynamic replication strategy based on a data classification model that would adapt the replication process according to user behavior towards data. This strategy dynamically and adaptively creates the replicas necessary in order to obtain the desired performance such as, reduced response time and improved system availability while ensuring the quality of service. The solution also attempts to meet customer requirements by respecting the SLA contract. The CloudSim simulator was used to evaluate the proposed strategy and compare it to other strategies. The results obtained showed an improvement in the criteria studied in a satisfactory manner.

In the near future, the Internet of NanoThings will enable the emerging of several unprecedented applications in several fields, which are until now unimaginable, and can not be realized with traditional communication networks. Software-Defined Metamaterials (SDMs) is a promising application recently proposed in the industrial field of smart materials, where a network of nanodevices is embedded in the structure of metamaterials, to allow the latter to change their electromagnetic behavior (e.g., cloaking, filtering and steering of sound and light) at runtime. Despite the routing schemes proposed for SDMs could cope with the unique challenges in the nanonetworks, such as the very high path loss and the extremely poor data computing and storing capabilities, there is no point-to-point routing scheme that directly considers the limited stored energy capabilities. The present work proposes three versions of an adjusted flood-based point-to-point routing scheme for static and dense 2D nanonetworks. These schemes aim to enhance the pioneering routing scheme proposed by Liaskos et al. in terms of energy efficiency while keeping high communication reliability. The results of extensive simulations over various performance scenarios using nano-sim tool on NS-3 show the advantages of the proposed schemes in terms of energy consumption, successful packet delivery ratio and forwarding packet rate.

An Internet of Things (IoT) network is a worldwide network connecting billions of devices with each other over the Internet. IPv6 Low power Wireless Personal Area Network (6LoWPAN) is an enabling technology for a complete end-to-end IoTs architecture. A 6LoWPAN contains small devices that communicate with other small devices or other powerful devices over the Internet. Communication privacy is a feature that allows the communicating users to make sure that there is no other entity that is spying on them. As such, it is a very important feature for the success of IoT applications. While traditionally privacy has been dealt with by hiding the content of the data content exchanged between the communicating pairs through the use of encryption, it has been shown that this cannot be sufficient in many application scenarios. It has been shown that metadata information such as the ones included in packet headers for the sake of the operation of communication protocols, such as IP addresses, reveal information such as the identities of the communicating pairs, which in some applications is considered critical information.

While there are many surveys in the literature dealing with communication privacy in the IoT, to our knowledge, little has been done on communication-flow identifiers privacy preservation in 6LoWPAN-based IoT networks. In this survey, we thoroughly expose the prime focus of the existing solutions on communication identifiers privacy in 6LoWPANs, clarifying the important information about: at which layer solutions operate, based on which protocol, against which attack, for which application, based on simulations or real prototypes, which sensitive information or communication identifiers are protected, which Privacy-Preserving Technique (PPT) is used, and how long is the duration of the protection against privacy attacks. We provide a comprehensive coverage on all of these aspects focusing on the main design guidelines that drove existing solutions while showing their merits and shortcomings.

Cloud computing is an effective technology that delivers interesting services to customers over the Internet. It is beneficial for scientific workflow systems in view of its powerful characteristics. However, scheduling workflow system over a cloud platform has become a challenging problem. In this paper, we propose a novel workflow scheduling strategy for the hybrid cloud environments which consists of an economical distribution of tasks between the various cloud service providers, in order to provide customers with high security services. Then, we study the impact of that security services on the total cost and deadline generated by the workflow. This problem is a major gap in the workflow scheduling field and it is still insufficiently explored in the literature. Our proposed scheduling system is composed for three modules. The first module is the Pre-Scheduler, the second is the Security Enhancement Module and the third one is the Post-Scheduler. The system evaluation and the extensive simulations are performed using an extension of Cloudsim simulation tool. The results show that our strategy preserves the same cost however it affects the deadline.

Routing in unmanned aerial vehicles (UAVs) faces serious energy consumption problems due to the premature battery depletion caused by the intensive exchange of control messages between the UAVs. The frequent use of control messages increases the delivery ratio. However, it increases the energy consumption which prevents the UAVs from completing their mission on time. In this paper, we present OEM-AODV, an adaptive routing protocol based on AODV, whereby the periodic control messages are dynamically sent to reduce energy consumption. We designed a predictive mathematical model through several steps based on performance modeling and multi-objective optimization using the Design of Experiments. Thereby, the frequency of sending periodic control messages is defined according to some mission parameters namely dimensions of the mission area, the maximum velocity of UAVs, and their transmission range. The simulation experiments using the NS-3 simulator show that OEM-AODV reduces significantly both energy consumption and routing overhead when compared with the original AODV and with a recent adaptive version EE-HELLO-AODV.

Diabetes is a chronic disease characterized by hyperglycemia where a person suffers from a high level of blood sugar, which leads to complications such as blindness, cardiovascular diseases, and amputation. It is expected that in 2040 the diabetic patients will reach 642 million globally. Hence considering this alarming figure there is a strong need to early diagnose and predict the symptoms of diabetes to save precious human lives. One possible way to diagnose this disease is to leverage machine learning algorithms. Machine learning has swiftly been infiltrating in various domains in healthcare. With the help of diabetes data, machine learning algorithms can find hidden patterns to predict whether a patient is diabetic or non-diabetic. This research aims to provide a comparative analysis of the performance and effectiveness of selected machine learning algorithms in predicting diabetes in women. We develop a predication framework and implemented ten different machine learning algorithms, namely: Naive Bayes, BayesNet, Decision Tree, Random Forest, AdaBoost, Bagging, K-Nearest Neighbor, Support Vector Machine, Logistic Regression, and Multi-Layer Perceptron. Experimental results procured for the Frankfurt hospital (Germany) dataset shows that K-Nearest Neighbor, Random Forest, and Decision Tree outperformed the other algorithms in terms of all metrics. We believe that our diabetes prediction framework will assist doctors to predict diabetes mellitus with high accuracy.

In mobile application stores, users very often rely on the opinions of others before downloading an application and its reputation could depend entirely on them. This makes analysis of users’ reviews very interesting for application owners to take future decisions. In this paper, we are interested in analyzing Algerian reviews on application store using sentiment analysis. To the best of our knowledge, this is the first study that explores the Algerian context where reviews have the particularity of being written using different languages (French, Arabic and Algerian Dialect) making them difficult to process. We analyzed these reviews according to two existing approaches: Automatic approach based on machine learning and Lexicon-based one. Evaluation of the proposed solution is conducted on more than 50 000 reviews collected from popular Algerian applications on Google play store. The obtained results are very promising, we achieved an accuracy of 80% using the lexicon-based approach and of 72% for SVM on Dialect reviews.

Learning ontological relations is an important step on the way to automatically developing ontologies. This paper introduces a novel way to exploit WordNet [16], the combination of pre-trained word embeddings and deep neural networks for the task of ontological relation classification. The data from WordNet and the knowledge encapsulated in the pre-trained word vectors are combined into an enriched dataset. In this dataset a pair of terms that are linked in WordNet through some ontological relation are represented by their word embeddings. A Deep Neural Network uses this dataset to learn the classification of ontological relations based on the word embeddings. The implementation of this approach has yielded encouraging results, which should help the ontology learning research community develop tools for ontological relation extraction.

Speech recognition is becoming increasingly used in real-world applications. One of the interesting applications is automatic gender recognition which aims to recognize male and female voices from short speech samples. This can be useful in applications such as automatic dialogue systems, system verification, prediction of demographic attributes (e.g., age, location) and estimating person’s emotional state. This paper focuses on gender identification from the publicly available dataset Arabic Natural Audio Dataset (ANAD) using an ensemble-classifier based approach. More specifically, initially we extended the original ANAD to include a gender label information through a manual annotation task. Next, in order to optimize the feature engineering process, a three stage machine learning approach is devised. In the first phase, re restricted to features to the two widely used ones; namely, MFCC and fundamental frequency coefficients. In the second phase, six distinct acoustic features were employed. Finally, in the third phase, the features were selected according to their associated weights in Random Forest Classifier, and the best features are thereby selected. The latter approach enabled us to achieve a classification rate of 96.02% on the test set generated with linear SVM classifier.

Facial recognition has always been a field of continuous development and research due to its usage in different areas such as security and robotics. It has gained even more popularity and interest by the researchers with the recent advancements in artificial intelligence and deep learning, which improved the robustness of facial recognition systems. In this paper, we focus on facial recognition using deep learning on small data sets with a limited number of individuals, for that we propose a local features based facial recognition approach that combines the robustness of feature extraction of CNN with the Harris corner detector. The experimental results of our proposed method surpassed the results of classical methods (LBP, Eigen Face, and Fisher Face) as well as recent works on Georgia Tech Face Database and AR Face Database and proved its efficiency and its robustness in different conditions including illumination variation, face pose variation, changes in facial expressions and face occlusions.

In this paper, we propose an efficient discrete wavelet transform-based compression method for color images. Generally, the strong correlation exists between the three planes R, G, and B of a color image, where the decrease of this correlation gives an improvement in the compression quality. The proposed method utilizes an efficient technique to reduce this correlation efficiently. In this regard, the main contribution is to design an optimized color space \(S_1S_2S_3\) using the PSO algorithm to represent the RGB image in a space more appropriate for performing the compression. The idea is to maximize the energy of the image in the plane \(S_1\) more than in \(S_2\) and \(S_3\). Moreover, we propose to optimize the thresholds appropriate for each plane of the converted image to partially reduce the number of the less important DWT coefficients that correspond to the lower quantity of energy. The obtained results facing those of state-of-the-art methods confirm that the proposed method shows clearly that the proposed method achieves high performances .

Facial Expression Recognition (FER) is a form of nonverbal communication; it translates the internal and emotional state of a human being by changing one or several facial muscles. Automated classification of facial expressions has known a great progress over the last decade; we observed the appearance of new methods based on Deep Learning (DL) instead of traditional classification methods. In this paper we propose an improved method based on Convolutional Neural Networks (CNN) that responds to the problem of classification of the six basic emotions(anger, disgust, fear, happy, sad and surprise) plus the neutral case. We validated our model on three public databases and we achieved better results than the state-of-the-art: CK+ 88,23%, JAFFE 86.24%, KDEF 82.38%. Our accuracies out perform results from recently proposed traditional methods as well as DL based methods.

A Data warehouse is a structure that stores big amount of data. This data is exploited in the best possible ways in order to improve the efficiency of decision-making. The huge volume of data makes answering queries complex and time-consuming. Therefore, materialized views are used in order to reduce the query processing time. Since materializing all views is not possible, due to space and maintenance constraints, materialized view selection became one of the crucial decisions in designing a data warehouse for optimal efficiency. In this paper, we propose a discrete - quantum based - version of Differential Evolution DE algorithm named QDE algorithm to solve the materialized view selection (MVS) problem with space constraint. This algorithm is a merging of the original DE with Quantum Evolutionary (QEA) algorithm. The experimental results show the efficiency of the proposed algorithm compared to well known algorithms used to solve MVS problem with space constraint such as HRUA and GEA.

Recommender systems are tools that provide personalized suggestions of items for users. They must be able to adapt constantly to user preferences and behavior in order to generate relevant recommendations. However, initial works in this field do not consider the context in which the users are. In recent years, a new recommendation technique, called Context-Aware Recommender System (CARS), has emerged. This approach integrates contextual information about users and/or items in the recommendation process to satisfy even more users’ needs. Therefore, accurate prediction depends upon the degree to which a recommendation method has incorporated the relevant contextual data. To address this issue, we propose to combine user based collaborative filtering with the Genetic Algorithm based meta-heuristic in order to provide better predictions for users. The proposed model uses a weighting function which incorporates the contextual factors that influence the users’ decisions. It is based on the Genetic Algorithm based meta-heuristic to estimate, for each contextual parameter, a degree of importance that would reduce the mean absolute error and increase the F-measure. Experimental results from Movielens dataset validate that our proposed algorithm improves recommendations accuracy.

Skin detection is a very important task in computer vision, since we can find it in many applications such as face detection and recognition, face tracking, gesture analysis, content-based image retrieval systems and human machine interaction systems. In this paper we present a novel rule-based skin detection method in the Cyan Magenta Yellow Key (CMYK) color space. This space is a subtractive color space used in color printing, and poorly explored in image processing and still less in skin detection tasks. Our method uses thresholds which are based on the relation between CMYK color components in order to recognize skin pixels, two thresholding models were proposed and we have considered the most performing one. The proposed method has been tested on two public skin image databases and has achieved very satisfactory qualitative and quantitative results against other widely used rule-based methods.

Most real-world problems are multiobjective in nature and considerable research efforts have been devoted to propose efficient multiobjective optimization approaches. Nondominated sorting genetic algorithm II (NSGA-II) is one of the well-known algorithms for this purpose which is based on a fast nondominated sorting procedure and an elitist selection strategy. This paper presents an improved NSGA-II algorithm (I-NSGA-II) based on greedy heuristics to tackle the minimum weight minimum connected dominating set problem. To make a trade-off between the size of the connected dominating set and its total weight, two objectives are considered, namely the minimization of the size and the minimization of the total edge-weight. The performance of I-NSGA-II is evaluated on a set of test problem instances with different sizes. Computational experiments show a significant improvement of our approach over NSGA-II with respect to the hypervolume indicator, run-time, and quality of solutions.

Ontology matching is a proper method to establish interoperability among heterogeneous ontologies. In this paper, we evaluate the ontology matching approach that we proposed previously and developed on a high level of accuracy. For that purpose, we performed a very detailed experimental study on six test cases of different domains from four OAEI tracks of various campaigns. The experimental results, adopting a cross-validation procedure and basing on standard evaluation measures, show a very high accuracy of matching. The proposed approach has proven its matching efficiency in front of all OAEI matching systems of the selected campaigns with major scores of all evaluation metrics adopted for all matching challenges. That permits to perfectly increase the performance of the ontology matching task.

Business processes modeling and verification are become essential to master and guarantee organization evolution. UML Activity diagrams have been used for this purpose, although they were not designed for that at the beginning. These diagrams lack for formal semantics which prohibits their verification. YAWL is a language developed for modeling workflows. It has a formal semantics based on Petri nets, and it is supported by open source toolsets. Transforming UML activity diagrams to YAWL is very beneficial. It allows profiting from the intuitiveness and widespread use of UML activity diagrams for modeling business processes and workflows, on one hand, and enabling their verification by using YAWL tools on the other hand.

Nowadays, manufacturing companies integrate several advanced technologies such as Internet of things, cyber-physical systems, smart factories, etc. to adapt themselves, as soon as possible, to the volatile market demands as well as to increase their throughput and profit. At the design level, the use of classical formal approaches such as Petri nets (either low or high-level) becomes neither convenient nor useful in the modeling and verification of modern manufacturing systems due to their rigid structure that cannot handle the reconfigurability of such systems. In this paper, we introduce configuration-dependent stochastic reward nets formalism that allows designers to study the reconfiguration in smart factories. The proposed formalism transforms configuration-dependent stochastic reward nets into stochastic reward nets in order to use the existing tools proposed for stochastic reward nets.