Ensemble Classification Approach for Cancer Prognosis and Prediction

Abstract

Gene expression data mostly available as cancer data have major challenges such as analyze, pattern matching and classification. Sometime task become more complex with large number of genes and small samples are available with noise and redundant information. Meaningful correlated information from dataset is the first and most important steps to be extracted for better diagnosis through artificial intelligence (AI). Accordingly, recent work for AI based classification and prognosis are focused in two steps process that is: (a) Feature extraction, and, (b) Ensemble Classification. Feature extraction will help in eliminating redundant and irrelevant genes, whereas ensemble classifier will help to optimize the accuracy. In this paper, we use double RBF kernel function for feature selection and novel fusion-procedure for enhance the performance of three base classifiers i.e., K Nearest Neighbor (KNN), Multi-Layer Perceptron (MLP) and Decision Tree (DT). Training of classifier is implemented based on k-fold cross validation techniques. The predicted accuracy of the proposed model has been compared with recent fusion methods such as Majority Voting, Distribution Summation and Dempster–Shafer on six benchmark cancer datasets. Experiment evaluation and result analysis gives promising and better performance than other fusion strategies, aiming at our goal-functions. Wisconsin Breast prognosis dataset is used with the proposed model for gene selection and prognosis prediction.