A Hybrid EM-Based Boosting Classification Model for Microarray Somatic Disease Prediction

Abstract

As the size of the micro-array disease databases increase, finding an essential feature set for the classification problem is complex due to the large data size and sparsity problems. Traditional feature subset models are based on static clustering and classification models due to the fixed sized dimensions cluster-based disease prediction process. Sparsity, missing values, and imbalance are the major issues that affect the selection of essential feature clusters for data classification process. In this chapter, a hybrid cluster-based Bayesian probability estimation model is proposed in order to predict the disease class label on high dimensional databases. The proposed cluster-based classification model selects optimal clusters for feature ranking and classification problems to improve the true positive rate and accuracy. Experimental results are simulated on different training datasets for accuracy prediction. The results proved that the gene-disease-based patterns have better optimization than the conventional methods in terms of statistical metrics and classification models.