Andre B. de Carvalho

A symbolic fault-prediction model based on multiobjective particle swarm optimization

In the literature the fault-proneness of classes or methods has been used to devise strategies for reducing testing costs and efforts. In general, fault-proneness is predicted through a set of design metrics and, most recently, by using Machine Learning (ML) techniques. However, some ML techniques cannot deal with unbalanced data, characteristic very common of the fault datasets and, their produced results are not easily interpreted by most programmers and testers. Considering these facts, this paper introduces a novel fault-prediction approach based on Multiobjective Particle Swarm Optimization (MOPSO). Exploring Pareto dominance concepts, the approach generates a model composed by rules with specific properties. These rules can be used as an unordered classifier, and because of this, they are more intuitive and comprehensible. Two experiments were accomplished, considering, respectively, fault-proneness of classes and methods. The results show interesting relationships between the studied metrics and fault prediction. In addition to this, the performance of the introduced MOPSO approach is compared with other ML algorithms by using several measures including the area under the ROC curve, which is a relevant criterion to deal with unbalanced data.

Non-Ordered Data Mining Rules Through Multi-Objective Particle Swarm Optimization: Dealing with Numeric and Discrete Attributes

Multi-objective meta-heuristics permit to conceive a complete novel approach to induce classifiers, where the properties of the rules can be expressed in different objectives, and then the algorithm finds these rules in an unique run by exploring Pareto dominance concepts. Furthermore, these rules can be used as an unordered classifier, in this way, the rules are more intuitive and easier to understand because they can be interpreted independently one of the other. This work describes a multi-objective particle swarm optimization (MOPSO) algorithm that handles with numerical and discrete attributes. The algorithm is evaluated by using the area under ROC curve and comparing the performance of the induced classifiers with other ones obtained with well known rule induction algorithms. The approximation sets produced by the algorithm are also analyzed following multi-objective methodology.

The Control of Dominance Area in Particle Swarm Optimization Algorithms for Many-Objective Problems

Multi-objective evolutionary algorithms (MOEA) are particulary suitable to solve real life problems, but they have some limitations when dealing with problems with many objectives, typically more than three. Recently, some many-objective techniques were proposed to avoid the deterioration of the search ability of Pareto dominance based MOEA for many-objective problems. This work applies the control of dominance area in two different Multi-objective Particle Swarm Optimization algorithms and investigates the influence of this technique in a cooperative-based framework. Besides, an empirical study is performed to identify if the many-objective technique increases the quality of the PSO algorithms for many-objective problems. The experimental results are compared applying some quality indicators and statistical test.

Mining Rules: A Parallel Multiobjective Particle Swarm Optimization Approach

Data mining is the overall process of extracting knowledge from data. In the study of how to represent knowledge in data mining context, rules are one of the most used representation form. However, the first issue in data mining is the computational complexity of the rule discovery process due to the huge amount of data. In this sense, this chapter proposes a novel approach based on a previous work that explores Multi-Objective Particle Swarm Optimization (MOPSO) in a rule learning context, called MOPSO-N. MOPSO-N applies MOPSO to search for rules with specific properties exploring Pareto dominance concepts. Besides, these rules can be used as an unordered classifier, so the rules are more intuitive and easier to understand because they can be interpreted independently one of the other. In this chapter, first some extensions to MOPSO-N are presented. These extensions are enhancements to the original algorithm to increase its performance, and to validate them, a wide set of experiments is conducted. Second, the main goal of this chapter, the parallel approach of MOPSO-N, called MOPSO-P, is described. MOPSO-P allows the algorithm to be applied to large datasets. The proposed MOPSO-P is evaluated, and the results showed that MOPSO-P is efficient for mining rules from large datasets.

Exploring multi-objective PSO and GRASP-PR for rule induction

A process for restructuring meat wherein chunked and wafer sliced meats are blended to promote release of natural binding proteins. The blended meats are formed into log shapes for freezing and tempering, and then pressed into shape and sliced into steaks or chops. The process produces a roast if the final slicing step is eliminated.

Analyzing the control of dominance area of solutions in particle swarm optimization for many-objective

The interest in the application of particle swarm optimization to solve different problems, especially multi-objective problems, grew in recent years. This metaheuristic is particularly suitable to solve real life problems, but like other multi-objective metaheuristics, has some limitations when dealing with problems with many objectives, typically more than three. Recently, some many-objective techniques were proposed to avoid the deterioration of the search ability of Pareto dominance based multi-objective evolutionary algorithms for many-objective problems. This work presents a study of the influence of the many-objective technique called the control of dominance area of solutions (CDAS) in multi-objective particle swarm optimization. It is presented an empirical analysis to identify the influence of the CDAS technique on the convergence and the diversity of a multi-objective PSO algorithm in many-objective scenarios through the analysis of some quality indicators and statistical tests.

A Non-ordered Rule Induction Algorithm through Multi-Objective Particle Swarm Optimization: Issues and Applications

Multi-Objective Metaheuristics permit to conceive a complete novel approach to induce classifiers, where the properties of the rules can be expressed in different objectives, and then the algorithm finds these rules in an unique run by exploring Pareto dominance concepts. Furthermore, these rules can be used as an unordered classifier, in this way, the rules are more intuitive and easier to understand because they can be interpreted independently one of the other. The quality of the learned rules is not affected during the learning process because the dataset is not modified, as in traditional rule induction approaches. With this philosophy, this chapter describes a Multi-Objective Particle Swarm Optimization (MOPSO) algorithm. One reason to choose the Particle Swarm Optimization Meta heuristic is its recognized ability to work in numerical domains. This propriety allows the described algorithm deals with both numerical and discrete attributes. The algorithm is evaluated by using the area under ROC curve and, by comparing the performance of the induced classifiers with other ones obtained with well known rule induction algorithms. The produced Pareto Front coverage of the algorithm is also analyzed following a Multi-Objective methodology. In addition to this, some application results in the Software Engineering domain are described, more specifically in the context of software testing. Software testing is a fundamental Software Engineering activity for quality assurance that is traditionally very expensive. The algorithm is used to induce rules for fault-prediction that can help to reduce testing efforts. The empirical evaluation and the comparison show the effectiveness and scalability of this new approach.