Binary classification of cancer microarray gene expression data using extreme learning machines

This paper presents the usage of Extreme Learning Machines for cancer microarray gene expression data. Extreme Learning Machines overcomes the problems of overfitting, local minima and improper training rate that are most common in traditional algorithms. We have evaluated the binary classification performance of Extreme Learning Machines on five bench marked datasets of cancer microarray gene expression data namely ALL/AML, CNS, Lung Cancer, Ovarian Cancer and Prostate Cancer. Feature Extraction has been performed using Correlation Coefficient prior to classification. The results indicate that ELM produces comparable or better results compared to the traditional classification methods like Naïve Bayes, Bagging, Random Forest and Decision Table.