Albert Orriols-Puig

Evolutionary rule-based systems for imbalanced data sets

This paper investigates the capabilities of evolutionary on-line rule-based systems, also called learning classifier systems (LCSs), for extracting knowledge from imbalanced data. While some learners may suffer from class imbalances and instances sparsely distributed around the feature space, we show that LCSs are flexible methods that can be adapted to detect such cases and find suitable models. Results on artificial data sets specifically designed for testing the capabilities of LCSs in imbalanced data show that LCSs are able to extract knowledge from highly imbalanced domains. When LCSs are used with real-world problems, they demonstrate to be one of the most robust methods compared with instance-based learners, decision trees, and support vector machines. Moreover, all the learners benefit from re-sampling techniques. Although there is not a re-sampling technique that performs best in all data sets and for all learners, those based in over-sampling seem to perform better on average. The paper adapts and analyzes LCSs for challenging imbalanced data sets and establishes the bases for further studying the combination of re-sampling technique and learner best suited to a specific kind of problem.

Fuzzy-UCS: A Michigan-Style Learning Fuzzy-Classifier System for Supervised Learning

This paper presents Fuzzy-UCS, a Michigan-style learning fuzzy-classifier system specifically designed for supervised learning tasks. Fuzzy-UCS is inspired by UCS, an on-line accuracy-based learning classifier system. Fuzzy-UCS introduces a linguistic representation of the rules with the aim of evolving more readable rule sets, while maintaining similar performance and generalization capabilities to those presented by UCS. The behavior of Fuzzy-UCS is analyzed in detail from several perspectives. The granularity of the linguistic fuzzy representation to define complex decision boundaries is illustrated graphically, and the test performance obtained with different inference schemes is studied. Fuzzy-UCS is also compared with a large set of other fuzzy and nonfuzzy learners, demonstrating the competitiveness of its on-line architecture in terms of performance and interpretability. Finally, the paper shows the advantages obtained when Fuzzy-UCS is applied to learn fuzzy models from large volumes of data.

Genetic-based machine learning systems are competitive for pattern recognition

During the last decade, research on Genetic-Based Machine Learning has resulted in several proposals of supervised learning methodologies that use evolutionary algorithms to evolve rule-based classification models. Usually, these new GBML approaches are accompanied by little experimentation and there is a lack of comparisons among different proposals. Besides, the competitiveness of GBML systems with respect to non-evolutionary, highly-used machine learning techniques has only been partially studied. This paper reviews the state of the art in GBML, selects some of the best representatives of different families, and compares the accuracy and the interpretability of their models. The paper also analyzes the behavior of the GBML approaches with respect to some of the most influential machine learning techniques that belong to different learning paradigms such as decision trees, support vector machines, instance-based classifiers, and probabilistic classifiers. The experimental observations emphasize the suitability of GBML systems for performing classification tasks. Moreover, the analysis points out the strengths of the different systems, which can be used as recommendation guidelines on which systems should be employed depending on whether the user prefers to maximize the accuracy or the interpretability of the models.

Facetwise Analysis of XCS for Problems With Class Imbalances

Michigan-style learning classifier systems (LCSs) are online machine learning techniques that incrementally evolve distributed subsolutions which individually solve a portion of the problem space. As in many machine learning systems, extracting accurate models from problems with class imbalances-that is, problems in which one of the classes is poorly represented with respect to the other classes-has been identified as a key challenge to LCSs. Empirical studies have shown that Michigan-style LCSs fail to provide accurate subsolutions that represent the minority class in domains with moderate and large disproportion of examples per class; however, the causes of this failure have not been analyzed in detail. Therefore, the aim of this paper is to carefully examine the effect of class imbalances on different LCS components. The analysis focuses on XCS, which is the most-relevant Michigan-style LCS, although the models could be easily adapted to other LCSs. Design decomposition is used to identify five elements that are crucial to guaranteeing the success of LCSs in domains with class imbalances, and facetwise models that explain these different elements for XCS are developed. All theoretical models are validated with artificial problems. The integration of all these models enables us to identify the sweet spot where XCS is able to scalably and efficiently evolve accurate models of rare classes; furthermore, facetwise analysis is used as a tool for designing a set of configuration guidelines that have to be followed to ensure convergence. When properly configured, XCS is shown to be able to solve highly unbalanced problems that previously eluded solution.

Bounding XCS's parameters for unbalanced datasets

This paper analyzes the behavior of the XCS classifier system on imbalanced datasets. We show that XCS with standard parameter settings is quite robust to considerable class imbalances. For high class imbalances, XCS suffers from biases toward the majority class. We analyze XCSs behavior under such extreme imbalances and prove that appropriate parameter tuning improves significantly XCSs performance. Specifically, we counterbalance the imbalance ratio by equalizing the reproduction probabilities of the most occurring and least occurring niches. The study provides guidelines to tune XCSs parameters for unbalanced datasets, based on the dataset imbalance ratio. We propose a method to estimate the imbalance ratio during XCSs training and adapt XCSs parameters online.

Revisiting UCS: Description, Fitness Sharing, and Comparison with XCS

This paper provides a deep insight into the learning mechanisms of UCS, a learning classifier system (LCS) derived from XCS that works under a supervised learning scheme. A complete description of the system is given with the aim of being useful as an implementation guide. Besides, we review the fitness computation, based on the individual accuracy of each rule, and introduce a fitness sharing scheme to UCS. We analyze the dynamics of UCS both with fitness sharing and without fitness sharing over five binary-input problems widely used in the LCSs framework. Also XCS is included in the comparison to analyze the differences in behavior between both systems. Results show the benefits of fitness sharing in all the tested problems, specially those with class imbalances. Comparison with XCS highlights the dynamics differences between both systems.

Large-Scale Experimental Evaluation of Cluster Representations for Multiobjective Evolutionary Clustering

Multiobjective evolutionary clustering algorithms are based on the optimization of several objective functions that guide the search following a cycle based on evolutionary algorithms. Their capabilities allow them to find better solutions than with conventional clustering algorithms if the suitable individual representation is selected. This paper provides a detailed analysis of the three most relevant and useful representations-prototype-based, label-based, and graph-based-through a wide set of synthetic data sets. Moreover, they are also compared to relevant conventional clustering algorithms. Experiments show that multiobjective evolutionary clustering is competitive with regard to other clustering algorithms. Furthermore, the best scenario for each representation is also presented.

Fuzzy knowledge representation study for incremental learning in data streams and classification problems

The extraction of models from data streams has become a hot topic in data mining due to the proliferation of problems in which data are made available online. This has led to the design of several systems that create data models online. A novel approach to online learning of data streams can be found in Fuzzy-UCS, a young Michigan-style fuzzy-classifier system that has recently demonstrated to be highly competitive in extracting classification models from complex domains. Despite the promising results reported for Fuzzy-UCS, there still remain some hot issues that need to be analyzed in detail. This paper carefully studies two key aspects in Fuzzy-UCS: the ability of the system to learn models from data streams where concepts change over time and the behavior of different fuzzy representations. Four fuzzy representations that move through the dimensions of flexibility and interpretability are included in the system. The behavior of the different representations on a problem with concept changes is studied and compared to other machine learning techniques prepared to deal with these types of problems. Thereafter, the comparison is extended to a large collection of real-world problems, and a close examination of which problem characteristics benefit or affect the different representations is conducted. The overall results show that Fuzzy-UCS can effectively deal with problems with concept changes and lead to different interesting conclusions on the particular behavior of each representation.

First approach toward on-line evolution of association rules with learning classifier systems

This paper presents CSar, a Michigan-style Learning Classifier System which has been designed for extracting quantitative association rules from streams of unlabeled examples. The main novelty of CSar with respect to the existing association rule miners isthat it evolves the knowledge on-line and so it is prepared to adapt its knowledge to changes in the variable associations hidden in the stream of unlabeled data quickly and efficiently. Preliminary results provided in this paper show that CSar is able to evolve interesting rules on problems that consist of both categorical and continuous attributes. Moreover, the comparison of CSar with Apriori on a problem that consists only of categorical attributes highlights the competitiveness of CSar with respect to more specific learners that perform enumeration to return all possible association rules. These promising results encourage us for further investigating on CSar.

Preliminary approach on synthetic data sets generation based on class separability measure

Usually, performance of classifiers is evaluated on real-world problems that mainly belong to public repositories. However, we ignore the inherent properties of these data and how they affect classifier behavior. Also, the high cost or the difficulty of experiments hinder the data collection, leading to complex data sets characterized by few instances, missing values, and imprecise data. The generation of synthetic data sets solves both issues and allows us to build problems with a minor cost and whose characteristics are predefined. This is useful to test system limitations in a controlled framework. This paper proposes to generate synthetic data sets based on data complexity. We rely on the length of the class boundary to build the data sets, obtaining a preliminary set of benchmarks to assess classifier accuracy. The study can be further matured to identify regions of competence for classifiers.