Population-Based Feature Selection for Biomedical Data Classification

Introduction

Data mining or knowledge discovery is a computational process of extracting hidden knowledge in large databases. The goal of data mining process is to extract useful information from a dataset. Figure 1 illustrates the phases of a data mining process. The first step in data mining process is to understanding of the problem. In the next step, data collect and prepare. In this step, data is cleaned from outlier instances or missing data and dataset reduces to only variables that are useful in a given data mining process. In the third step, a mining model or model is built. The quality of a model can evaluate using a number of the techniques. The last step in the data mining process is to deploy the models to a real environment.

Figure 1.

The data mining process

Data mining techniques have been successfully used in various biomedical domains, for example the detection of tumors, the diagnosis of cancers and other diseases. One of the main challenge in biomedical data mining and analysis is the so called “curse of dimensionality”. Especially the biomedical data are presented by relatively few instances and exhibited in a high dimensional feature space(Peng, Wu, & Jiang, 2010). Feature selection, a process in data transformation phase, reduces the number of features, removes irrelevant, redundant and misleading features, which leads to expediting learning algorithm and improves predictive performance. Feature selection algorithms are divided into three categories: wrapper methods that uses the learning algorithms to evaluate the usefulness of features, filter methods that evaluate features according to the statistical characteristics of the data, and embedded methods that feature selection embed in the learning algorithm. Population based metaheuristics such as genetic algorithm, particle swarm optimization, Imperialist competitive algorithm, artificial bee algorithm, Ant colony optimization, and leap frog optimization have been considered as effective wrapper feature selection approach. These metaheuristics are based on a population of solutions and an iterative procedure. At each iteration, they try to find a better solution than previous solutions using some operators. Feature selection algorithms have been successfully applied in various biomedical domains. A. Antoniadis et al, (2003) presented a statistical feature reduction approach for the classification of tumors. I. Guyan et al, (2002) address the problem of selection of a small subset of genes from broad patterns of gene expression data, recorded on DNA micro-arrays. Using available training examples from cancer and normal patients, they build a classifier suitable for genetic diagnosis, as well as drug discovery. In another work, wrapper approaches was applied for gene selection(Blanco, Larrañaga, Inza, & Sierra, 2004). Y. Peng et al. (2010) presents a novel feature selection approach to deal with issues of high dimensionality in biomedical data classification. The approach proposed in this paper integrated filter and wrapper methods into a sequential search procedure with the aim to improve the classification performance of the features selected. In this chapter, we focus on application of population based feature selection algorithms for biomedical data classification. To this purpose, four population based metaheuristics are considered: genetic algorithm, particle swarm optimization, artificial bee algorithm, and imperialist competitive algorithm. We also analyze the efficiency of this approach on four biomedical dataset: Wisconsin Diagnostic Breast Cancer, Wisconsin Prognostic Breast Cancer, SPECTF heart dataset, and Hepatitis diagnosis.