Anne M. P. Canuto

Investigating the influence of the choice of the ensemble members in accuracy and diversity of selection-based and fusion-based methods for ensembles

One of the most important steps in the design of a multi-classifier system (MCS), also known as ensemble, is the choice of the components (classifiers). This step is very important to the overall performance of a MCS since the combination of a set of identical classifiers will not outperform the individual members. The ideal situation would be a set of classifiers with uncorrelated errors - they would be combined in such a way as to minimize the effect of these failures. This paper presents an extensive evaluation of how the choice of the components (classifiers) can affect the performance of several combination methods (selection-based and fusion-based methods). An analysis of the diversity of the MCSs when varying their components is also performed. As a result of this analysis, it is aimed to help designers in the choice of the individual classifiers and combination methods of an ensemble.

A comparative analysis of genetic algorithm and ant colony optimization to select attributes for an heterogeneous ensemble of classifiers

In the context of ensemble systems, feature selection methods can be used to provide different subsets of attributes for the individual classifiers, aiming to reduce redundancy among the attributes of a pattern and to increase the diversity in such systems. Among the several techniques that have been proposed in the literature, optimization methods have been used to find the optimal subset of attributes for an ensemble system. In this paper, an investigation of two optimization techniques, genetic algorithm and ant colony optimization, will be used to guide the distribution of the features among the classifiers. This analysis will be conducted in the context of heterogeneous ensembles and using different ensemble sizes.

A Dynamic Classifier Selection Method to Build Ensembles using Accuracy and Diversity

Ensemble of classifiers is an effective way of improving performance of individual classifiers. However, the choice of the ensemble members can become a very difficult task, in which, in some cases, it can lead to ensembles with no performance improvement. In order to avoid this situation, there is a need to find effective classifier member selection methods. In this paper, a DCS (Dynamic Classifier Selection)-based method is presented, which takes into account performance and diversity of the classifiers in order to choose the ensemble members.

Filter-based optimization techniques for selection of feature subsets in ensemble systems

Feature selection methods select a subset of attributes (features) of a dataset and it is done based on a defined measure, eliminating the redundant and irrelevant ones. When a feature selection method is applied in a dataset, we aim to improve the quality of the dataset representation. For ensemble systems, feature selection techniques can supply different feature subsets for the individual components, reducing the redundancy that can exist among the features of an input pattern and to increase the diversity level of these systems. This paper proposes the application of three well-known optimization techniques (particle swarm optimization, ant-colony optimization and genetic algorithms), in both mono and bi-objective versions, to choose subsets of features for the individual components of ensembles. The feature selection process was based on two filter-based evaluation criteria that tried to capture the idea of diversity of individual classifiers and group diversity of an ensemble system. In this case, these optimization techniques try to maximize these diversities measures, either individually (mono-objective) or together (bi-objective). An empirical analysis was performed, where all ensemble systems were applied to 11 datasets and we compared both mono and bi-objective versions among each other and with a random subset procedure. Based on the empirical analysis, we will observe that PSO with a bi-objective function will be the most promising direction, when selecting attributes for individual components of ensemble systems.

Investigating fusion approaches in multi-biometric cancellable recognition

Highlights? We analyzed fusion approaches for cancellable multi-biometric data, with ensembles. ? We adapted three transformation functions (FTs) to be used with voice and iris data. ? We find that individual FTs decrease the accuracy of the voice and iris dataset. ? The combination of transformation functions increased the accuracy of the ensembles. ? The statistical analysis proved the good results reached by combining all three FTs. Cancellable biometrics has recently been introduced in order to overcome some privacy issues about the management of biometric data, aiming to transform a biometric trait into a new but revocable representation for enrolment and identification (verification). Therefore, a new representation of original biometric data can be generated in case of being compromised. Additionally, the use multi-biometric systems are increasingly being deployed in various biometric-based applications since the limitations imposed by a single biometric model can be overcome by these multi-biometric recognition systems. In this paper, we specifically investigate the performance of different fusion approaches in the context of multi-biometrics cancellable recognition. In this investigation, we adjust the ensemble structure to be used for a biometric system and we use as examples two different biometric modalities (voice and iris data) in a multi-biometrics context, adapting three cancellable transformations for each biometric modality.

Using Accuracy and Diversity to Select Classifiers to Build Ensembles

Ensemble of classifiers is an effective way of improving performance of individual classifiers. However, the task of selecting the ensemble members is often a non-trivial one. For example, in some cases, a bad selection strategy could lead to ensembles with no performance improvement. Thus, many researchers have put a lot of effort in finding an effective method for selecting classifier for building ensembles. In this context, a dynamic classifier selection (DCS) method is proposed, which takes into account both the accuracy and the diversity of the classifiers.

Use of multi-objective genetic algorithms to investigate the diversity/accuracy dilemma in heterogeneous ensembles

Classifier ensembles, also known as committees, are systems composed of a set of base classifiers (organized in a parallel way) and a combination module, which is responsible for providing the final output of the system. The main aim of using ensembles is to provide better performance than the individual classifiers. In order to build robust ensembles, it is often required that the base classifiers are as accurate as diverse among themselves - this is known as the diversity/accuracy dilemma. There are, in the literature, some works analyzing the ensemble performance in context of such a dilemma. However, the majority of them address the homogenous structures, i.e., ensembles composed only of the same type of classifiers. Motivated by such a limitation, this paper presents an empirical investigation on the diversity/accuracy dilemma for heterogeneous ensembles. In order to do so, multi-objective genetic algorithms will be used to guide the building of the ensemble systems.

An Investigation of the Effects of Variable Vigilance within the RePART Neuro-Fuzzy Network

RePART is a variation of fuzzy ARTMAP to which a reward/punishment concept has been added. Previously, an improvement in performance of RePART had been noted compared with other ARTMAP-based models, such as fuzzy ARTMAP and ARTMAP-IC. In this paper, a wider investigation of RePART performance is described, in which RePART is analysed in relation to a multi-layer perceptron and a RAM-based network in a handwritten numeral recognition task. In the RePART network, a variable vigilance parameter is proposed in order to smooth the poor-generalisation problem of RePART. Firstly, the same vigilance is associated within every neuron – general variable vigilance. Secondly, an individual variable vigilance for each neuron – which takes into account its average and frequency of activation – is used. In a handwritten numeral recognition task using individual variable vigilance, RePART performance improved and demonstrated a performance comparable with alternative architectures such as fuzzy multi-layer perceptron and Radial RAM.

Analyzing the Performance of an Agent-based Neural System for Classification Tasks Using Data Distribution among the Agents

The use of intelligent agents in the structure of multi-classifier systems has been investigated in order to overcome some drawbacks of these systems and, as a consequence, to improve the performance of such systems. As a result of this, the NeurAge system was proposed. This system has presented good results in some conventional (centralized) classification tasks. Nevertheless, in some classification tasks, relevant features can be distributed over a set of agent. These applications can be classified as distributed classification tasks. In this paper, an investigation of the performance of the NeurAge system in distributed classification tasks will be performed. In other words, it will be investigated the performance of NeurAge when the data are distributed over its agents.

Carcara: A Multi-agent System for Web Mining Using Adjustable User Profile and Dynamic Grouping

This paper presents Caracara, a system for searching and mining information on the World Wide Web, using adjustable user profiles and a dynamic grouping process. Suggestions for visiting URLs will be made to users, according to the user profiles. Also, the system has a dynamic strategy of suggestions (dynamic grouping), which helps the cooperation among the agents as well as the distribution of information. All the needed information to define the user profile is acquired, in a non-invasive way, through monitoring the user during its use of the system.