A Comparative Performance Evaluation of Random Forest Feature Selection on Classification of Hepatocellular Carcinoma Gene Expression Data

Abstract

Hepatocellular carcinoma is one of the cancers that cause death in the world. We get hepatocellular carcinoma data in the form of microarray data gene expression obtained from the National Center for Biotechnology Information website consisting of 40 samples and 54675 features. The main purpose of this research is to compare the performance evaluation of Hepatocellular Carcinoma by applying feature selection to several classification algorithms. Random Forest feature selection method will be paired with several classification algorithms such as Support Vector Classification, Neural Network Classification, Random Forest, Logistic Regression, and Naïve Bayes. This study uses 5-fold cross-validation as an evaluation method. The results showed that Random Forest algorithm, Neural Network, Vector Machine Classification, and Naive Bayes show higher classification performance evaluation than without using random forest feature selection, while the Logistic Regression model provides a higher performance evaluation without using Random Forest feature selection. Support Vector Classification offers the highest performance evaluation compared to four other algorithms using feature selection, but Logistic Regression provides higher performance evaluation compared to different classification algorithms without feature selection.