Arnaud Quirin

A Study on the Use of Multiobjective Genetic Algorithms for Classifier Selection in FURIA-based Fuzzy Multiclassifiers

Abstract In a preceding contribution, we conducted a study considering a fuzzy multiclassifier system (MCS) design framework based on Fuzzy Unordered Rule Induction Algorithm (FURIA). It served as the fuzzy rule classification learning algorithm to derive the component classifiers considering bagging and feature selection. In this work, we integrate this approach under the overproduce-and-choose strategy. A state-of-the-art evolutionary multiobjective algorithm, namely NSGA-II, is used to provide a component classifier selection and improve FURIA-based fuzzy MCS. We propose five different fitness functions based on three different optimization criteria, accuracy, complexity, and diversity. Twenty UCI high dimensional datasets were considered in order to conduct the experiments. A combination between accuracy and diversity criteria provided very promising results, becoming competitive with classical MCS learning methods.

A Genetic Fuzzy Linguistic Combination Method for Fuzzy Rule-Based Multiclassifiers

Fuzzy set theory has been widely and successfully used as a mathematical tool to combine the outputs provided by the individual classifiers in a multiclassification system by means of a fuzzy aggregation operator. However, to the best of our knowledge, no fuzzy combination method has been proposed, which is composed of a fuzzy rule-based system. We think this can be a very promising research line as it allows us to benefit from the key advantage of fuzzy systems, i.e., their interpretability. By using a fuzzy linguistic rule-based classification system as a combination method, the resulting classifier ensemble would show a hierarchical structure, and the operation of the latter component would be transparent to the user. Moreover, for the specific case of fuzzy multiclassification systems, the new approach could also become a smart way to allow standard fuzzy classifiers to deal with high-dimensional problems, avoiding the curse of dimensionality, as the chance to perform classifier selection at class level is also incorporated, into the method. We conduct comprehensive experiments considering 20 UCI datasets with different dimensionality, where our approach improves or at least maintains accuracy, while reducing complexity of the system, and provides some interpretability insight into the multiclassification system reasoning mechanism. The results obtained show that this approach is able to compete with the state-of-the-art multiclassification system selection and fusion methods in terms of accuracy, thus providing a good interpretability-accuracy tradeoff.

Debugging complex software systems by means of pathfinder networks

This paper introduces a new methodology based on the use of Pathfinder networks (PFNETs) for the debugging of multi-agent systems (MASs). This methodology is specifically designed to develop a forensic analysis (i.e. a debugging process performed on previously recorded data of the MAS run) of MASs showing complex tissues of relationships between agents (i.e. a high complexity in their social level). Like previous works in the field of forensic analysis of MASs, our approach is performed by considering displays of the system activity which aim to be understandable by human beings. These displays allow us to understand the social behavior of the system, discover emergent behaviors, and debug possible undesirable behaviors. However, it is well known that the visualization of information in a humanly comprehensible way becomes a complex task when large amounts of information have to be represented, as is the case of the social behavior of large-scale MASs. Our methodology tackles this problem through the use of PFNETs, which are considered to reduce the data complexity in order to obtain simple representations that show only the most important global interactions in the system. In addition, the proposed methodology is customizable thanks to the use of two thresholds allowing the user to define the desired specificity level in the display. The proposal is illustrated with a detailed case study considering a complex customer-seller MAS.

Comparing two genetic overproduce-and-choose strategies for fuzzy rule-based multiclassification systems generated by bagging and mutual information-based feature selection

In [14] we proposed a scheme to generate fuzzy rule-based multiclassification systems by means of bagging, mutual information-based feature selection, and a multicriteria genetic algorithm for static component classifier selection guided by the ensemble training error. In the current contribution we extend the latter component by making use of the bagging approachs capability to evaluate the accuracy of the classifier ensemble using the out-of-bag estimates. An exhaustive study is developed on the potential of the two multicriteria genetic algorithms respectively considering the classical training error and the out-of-bag error fitness functions to design a final multiclassifier with an appropriate accuracy-complexity trade-off. Several parameter settings for the global approach are tested when applied to nine popular UCI datasets with different dimensionality.

MOSubdue: a Pareto dominance-based multiobjective Subdue algorithm for frequent subgraph mining

Graph-based data mining approaches have been mainly proposed to the task popularly known as frequent subgraph mining subject to a single user preference, like frequency, size, etc. In this work, we propose to deal with the frequent subgraph mining problem from multiobjective optimization viewpoint, where a subgraph (or solution) is defined by several user-defined preferences (or objectives), which are conflicting in nature. For example, mined subgraphs with high frequency are often of small size, and vice-versa. Use of such objectives in the multiobjective subgraph mining process generates Pareto-optimal subgraphs, where no subgraph is better than another subgraph in all objectives. We have applied a Pareto dominance approach for the evaluation and search subgraphs regarding to both proximity and diversity in multiobjective sense, which has incorporated in the framework of Subdue algorithm for subgraph mining. The method is called multiobjective subgraph mining by Subdue (MOSubdue) and has several advantages: (i) generation of Pareto-optimal subgraphs in a single run (ii) selection of subgraph-seeds from the candidate subgraphs based on all objectives (iii) search in the multiobjective subgraphs lattice space, and (iv) capability to deal with different multiobjective frequent subgraph mining tasks by customizing the tackled objectives. The good performance of MOSubdue is shown by performing multiobjective subgraph mining defined by two and three objectives on two real-life datasets.

Graph-based data mining: A new tool for the analysis and comparison of scientific domains represented as scientograms

The creation of some kind of representations depicting the current state of Science (or scientograms) is an established and beaten track for many years now. However, if we are concerned with the automatic comparison, analysis and understanding of a set of scientograms, showing for instance the evolution of a scientific domain or a face-to-face comparison of several countries, the task is titanically complex as the amount of data to analyze becomes huge and complex. In this paper, we aim to show that graph-based data mining tools are useful to deal with scientogram analysis. Subdue, the first algorithm proposed in the graph mining area, has been chosen for this purpose. This algorithm has been customized to deal with three different scientogram analysis tasks regarding the evolution of a scientific domain over time, the extraction of the common research categories substructures in the world, and the comparison of scientific domains between different countries. The outcomes obtained in the developed experiments have clearly demonstrated the potential of graph mining tools in scientogram analysis.

A multiobjective evolutionary programming framework for graph-based data mining

Subgraph mining is the process of identifying concepts describing interesting and repetitive subgraphs within graph-based data. The exponential number of possible subgraphs makes the problem very challenging. Existing methods apply a single-objective subgraph search with the view that interesting subgraphs are those capable of not merely compressing the data, but also enhancing the interpretation of the data considerably. Usually the methods operate by posing simple constraints (or user-defined thresholds) such as returning all subgraphs whose frequency is above a specified threshold. Such search approach may often return either a large number of solutions in the case of a weakly defined objective or very few in the case of a very strictly defined objective. In this paper, we propose a framework based on multiobjective evolutionary programming to mine subgraphs by jointly maximizing two objectives, support and size of the extracted subgraphs. The proposed methodology is able to discover a nondominated set of interesting subgraphs subject to tradeoff between the two objectives, which otherwise would not be achieved by the single-objective search. Besides, it can use different specific multiobjective evolutionary programming methods. Experimental results obtained by three of the latter methods on synthetically generated as well as real-life graph-based datasets validate the utility of the proposed methodology when benchmarked against classical single-objective methods and their previous, nonevolutionary multiobjective extensions.

Social Network Analysis of Co-fired Fuzzy Rules

The popularity of modern online social networks has grown up so quickly in the last few years that, nowadays, social network analysis has become one of the hottest research lines in the world. It is important to highlight that social network analysis is not limited to the analysis of networks connecting people. Indeed, it is strongly connected with the classical methods widely recognized in the context of graph theory. Thus, social network analysis is applied to many different areas like for instance economics, bibliometrics, and so on. This contribution shows how it can also be successfully applied in the context of designing interpretable fuzzy systems. The key point consists of looking at the rule base of a fuzzy system as a fuzzy inference-gram (fingram), i.e., as a social network made of nodes representing fuzzy rules. In addition, nodes are connected through edges that represent the interaction between rules, at inference level, in terms of co-fired rules, i.e., rules fired at the same time by a given input vector. In short, fingram analysis consists of studying the interaction among nodes in the network for the purpose of understanding the structure and behavior of the fuzzy rule base under consideration. It is based on the basic principles of social network analysis which have been properly adapted to the design of fuzzy systems.

On the Use of Bagging, Mutual Information-Based Feature Selection and Multicriteria Genetic Algorithms to Design Fuzzy Rule-Based Classification Ensembles

In this contribution we explore the combination of bagging with random subspace and two variants of Battitis mutual information feature selection methods to design fuzzy rule-based classification system ensembles. Besides, we consider a multicriteria genetic algorithm guided by the training error to select the component classifiers, in order to look for appropriate accuracy-complexity trade-offs in the final multiclassifier.

A multiobjective variant of the Subdue graph mining algorithm based on the NSGA-II selection mechanism

In this work we propose a Pareto-based multi-objective search strategy for subgraph mining in structural databases. The method is an extension of Subdue, a classical graph-based knowledge discovery algorithm, and it is thus called MultiObjective Subdue (MOSubdue). MOSubdue incorporates the NSGA-IIs crowding selection mechanism in order to retrieve a well distributed Pareto optimal set of meaningful subgraphs showing different optimal trade-offs between support and complexity, in a single run. The good performance of the proposed approach is empirically demonstrated by using a reallife data set concerning the analysis of web sites.