Feature Selection Techniques for Bioinformatics Data Analysis

The emergence of microarray datasets sparked a new branch of bioinformatics research. An important problem in microarray research is the selection of true biomarkers (a most relevant feature subset) for disease identification/classification using gene expression data. This problem has received the most interest in the context of cancer research to date. In this work we have experimented with various machine learning techniques for true biomarker selection for cancer classification. This research work carried on two bioinformatics datasets with application of six feature selection algorithms. Major contribution of the work is to identify these top performing efficient feature selection algorithms through an exhaustive survey and then implementing ensemble learning technique using these algorithms. A novel contribution of this research work is to implement feature weighting as an ensemble operator and compare efficiency of feature set selection using this approach with the others. The experimental results revealed that the proposed feature weighting as an ensemble operator can choose feature sets efficiently.