International Journal on Artificial Intelligence Tools

The naïve Bayes model is a simple but often satisfactory supervised classification method. The original naïve Bayes scheme, does, however, have a serious weakness, namely, the harmful effect of redundant predictors. In this paper, we study how to apply a regularization technique to learn a computationally efficient classifier that is inspired by naïve Bayes. The proposed formulation, combined with an L₁-penalty, is capable of discarding harmful, redundant predictors. A modification of the LARS algorithm is devised to solve this problem. We tackle both real-valued and discrete predictors, assuring that our method is applicable to a wide range of data. In the experimental section, we empirically study the effect of redundant and irrelevant predictors. We also test the method on a high dimensional data set from the neuroscience field, where there are many more predictors than data cases. Finally, we run the method on a real data set than combines categorical with numeric predictors. Our approach is compared with several naïve Bayes variants and other classification algorithms (SVM and kNN), and is shown to be competitive.

Collaborative learning environments require intensive, regular and frequent analysis of the increasing amount of interaction data generated by students to assess that collaborative learning takes place. To support timely assessments that may benefit students and teachers the method of analysis must provide meaningful evaluations while the interactions take place. This research proposes machine learning-based techniques to infer the relationship between student collaboration and some quantitative domain-independent statistical indicators derived from large-scale evaluation analysis of student interactions. This paper (i) compares a set of metrics to identify the most suitable to assess student collaboration, (ii) reports on student evaluations of the metacognitive tools that display collaboration assessments from a new collaborative learning experience and (iii) extends previous findings to clarify modeling and usage issues. The advantages of the approach are: (1) it is based on domain-independent and generally observable features, (2) it provides regular and frequent data mining analysis with minimal teacher or student intervention, thereby supporting metacognition for the learners and corrective actions for the teachers, and (3) it can be easily transferred to other e-learning environments and include transferability features that are intended to facilitate its usage in other collaborative and social learning tools.

Although the importance of trust in recommender systems is widely recognized, the actual mechanisms of trust propagation and trust preservation are poorly understood. This is partly due to the fact that trust is a complex notion, which is typically context dependent, subjective, dynamic and not always transitive or symmetrical. This paper presents a theoretical analysis of the notion of trust in news recommendation and discusses the advantages of modeling this notion using Defeasible Logic Programming, a general-purpose defeasible argumentation formalism based on logic programming. In the proposed framework, users can express explicit trust statements on news reports, news sources and other users. Trust is then modeled and propagated using a dialectical process supported by a Defeasible Logic Programming interpreter. A set of basic postulates for trust and their representation by means of defeasible rules is presented. The suitability of the approach is investigated with a set of illustrative examples and then analyzed from a formal perspective. The obtained results indicate that the proposed framework provides a solid foundation for building trust-based news recommendation services.

In the past few years, evolutionary algorithm ensembles have gradually attracted more and more attention in the community of evolutionary computation. This paper proposes a novel evolutionary algorithm ensemble for global numerical optimization, named NEALE. In order to make a good tradeoff between the exploration and exploitation, NEALE is composed of two constituent algorithms, i.e., the composite differential evolution (CoDE) and the covariance matrix adaptation evolution strategy (CMA-ES). During the evolution, CoDE aims at probing more promising regions and refining the overall quality of the population, while the purposes of CMA-ES are to accelerate the convergence speed and to enhance the accuracy of the solutions. In addition, NEALE encourages the interaction between the constituent algorithms. In NEALE, the interaction is controlled by a predefined generation number and different interaction strategies are designed according to the features of the constituent algorithms. The performance of NEALE has been tested on 25 benchmark test functions developed for the special session on real-parameter optimization of the 2005 IEEE Congress on Evolutionary Computation (IEEE CEC2005). Compared with other state-of-the-art evolutionary algorithms and the individual constituent algorithms, NEALE performs significantly better than them.

In this paper we develop a new hybrid algorithm incorporating the penalty function technique for solving nonlinear constrained optimization problems. The principle is based on converting the constrained optimization problem into an unconstrained optimization problem by the penalty function technique. Then, we have proposed a new penalty technique, called Big-M penalty that is different from the existing ones. Accordingly, a hybrid algorithm has been developed based on Split and Discard Strategy (SDS) and advanced real coded genetic algorithm (ARCGA), with tournament selection, multiparent whole arithmetical crossover, double mutation (boundary and whole nonuniform mutation) and elitism. In SDS technique, the entire search space is divided into two equal subregions. Then the one containing the feasible solution with better fitness value is selected. This process is repeated until the accepted subregion reduces to a very small region with negligible edges. Finally, to test the performance of the proposed method along with three different penalty function techniques, it is applied to several well-known benchmark test problems available in the literature.

Feature selection is a multi-objective problem with the two main conflicting objectives of minimising the number of features and maximising the classification performance. However, most existing feature selection algorithms are single objective and do not appropriately reflect the actual need. There are a small number of multi-objective feature selection algorithms, which are wrapper based and accordingly are computationally expensive and less general than filter algorithms. Evolutionary computation techniques are particularly suitable for multi-objective optimisation because they use a population of candidate solutions and are able to find multiple non-dominated solutions in a single run. However, the two well-known evolutionary multi-objective algorithms, non-dominated sorting based multi-objective genetic algorithm II (NSGAII) and strength Pareto evolutionary algorithm 2 (SPEA2) have not been applied to filter based feature selection. In this work, based on NSGAII and SPEA2, we develop two multi-objective, filter based feature selection frameworks. Four multi-objective feature selection methods are then developed by applying mutual information and entropy as two different filter evaluation criteria in each of the two proposed frameworks. The proposed multi-objective algorithms are examined and compared with a single objective method and three traditional methods (two filters and one wrapper) on eight benchmark datasets. A decision tree is employed to test the classification performance. Experimental results show that the proposed multi-objective algorithms can automatically evolve a set of non-dominated solutions that include a smaller number of features and achieve better classification performance than using all features. NSGAII and SPEA2 outperform the single objective algorithm, the two traditional filter algorithms and even the traditional wrapper algorithm in terms of both the number of features and the classification performance in most cases. NSGAII achieves similar performance to SPEA2 for the datasets that consist of a small number of features and slightly better results when the number of features is large. This work represents the first study on NSGAII and SPEA2 for filter feature selection in classification problems with both providing field leading classification performance.

We present a new ensemble learning method that employs a set of regional classifiers, each of which learns to handle a subset of the training data. We split the training data and generate classifiers for different regions in the feature space. When classifying an instance, we apply a weighted voting scheme among the classifiers that include the instance in their region. We used 11 datasets to compare the performance of our new ensemble method with that of single classifiers as well as other ensemble methods such as RBE, bagging and Adaboost. As a result, we found that the performance of our method is comparable to that of Adaboost and bagging when the base learner is C4.5. In the remaining cases, our method outperformed other approaches.

Overworking behaviors appear in multi-agent systems specially when there are unbalanced communication patterns. This occurs when an agent receives many messages in a short period of time in comparison to other agents. As the agent pays attention to the large amount of messages, it worsens its performance, causing the system's performance to worsen as well, while other agents with similar services can be idle. The reasons for these behaviors are varied and depend on the nature of the messages. This article presents a measurement approach that detects unbalanced overworking situations in multi-agent systems with a new metric. Moreover, this approach includes a set of recommendations that determine possible common causes of this defective behavior depending on the relations of different measurement values. These recommendations also propose a solution for repairing each cause, suppressing these unbalanced overworking situations. A tool has been developed for allowing designers to measure the agents' communications, to obtain the diagnosis, and to know the recommended solutions. The current work has been applied in two case studies, whose results advocate that the suppression of unbalanced overworking situations is strongly related to the improvement of performance in multi-agent systems. Furthermore, the experience of this approach in nine different problem domains is presented as support for the causes and solutions of the set of recommendations.

Feature selection is essential in data mining and pattern recognition, especially for database classification. During past years, several feature selection algorithms have been proposed to measure the relevance of various features to each class. A suitable feature selection algorithm normally maximizes the relevancy and minimizes the redundancy of the selected features. The mutual information measure can successfully estimate the dependency of features on the entire sampling space, but it cannot exactly represent the redundancies among features. In this paper, a novel feature selection algorithm is proposed based on maximum relevance and minimum redundancy criterion. The mutual information is used to measure the relevancy of each feature with class variable and calculate the redundancy by utilizing the relationship between candidate features, selected features and class variables. The effectiveness is tested with ten benchmarked datasets available in UCI Machine Learning Repository. The experimental results show better performance when compared with some existing algorithms.