Hybrid Gene Selection Framework using Adaptive Wrapper and Filtering Techniques

Abstract

Analysing the thousands of gene expression values is a difficult task due to the curse of dimensionality of data produced by Microarray chips. Primary role of an effective feature selection model is to simplify this task. To simplify the task of disease classification and predicting cancer, feature selection plays a vital role through removing less informative genes. In this study, we propose a hybrid approach to gene selection using adaptive filter and adaptive wrapper approach. As filter method exhibits some limitations, an adaptive form of filtering has been employed that iteratively selects genes in each iteration and emphasizes on the misclassified samples and in subsequent iteration tries to find out the effective genes for misclassified samples. This approach performs better than traditional filter methods as it focuses on its weaknesses. In gene selection, Artificial Neural Networks (ANN) are mostly used as a classifier. In this study, adaptive ANN has been used as an internal wrapper. This helps to generate a better subset of genes. The proposed hybrid approach is applied on leukaemia, colon and lung cancer benchmarked datasets. Better result has been found compared to other well-known approaches.