MULTI FILTER ENSEMBLE METHOD FOR CANCER PROGNOSIS AND DIAGNOSIS

The nature of the gene expression profiles are high dimension, very small sample size, continuous types so it is really a challenged task to achieve good classification accuracy from the tumor samples. The main aim of feature selection is to find out most relevant features, which may increase the computational speed and accuracy. We thus proposed a multi filter ensemble based hybrid gene selection method. Here we have used four filter methods such as Information Gain (IG), Gain ratio (GR), Relief, Correlation to filter the irreverent and redundant genes. By the help of computationally efficient filters candidate features are selected .The top N genes with highest rank of individual subset are integrated to produce a new dataset. Then SVM attribute evaluator is applied for attribute evaluation purpose. Finally LIBSVM classifier is used to detect the best feature subset. This experimental result proves the proposed method is quite efficient then other gene selection methods and it provide a high accuracy under some characteristics genes. Keywords— IG, GR, CORRELATION, ReliefF, LIBSVM