Classification of Gene Expression Data Using Ensemble Learning with Effective Feature Selection and Optimization Techniques

Cancer is the most lethal disease impacting human life globally. When identified at an early stage, it is possible to treat the disease effectively during its onset. Nowadays, early cancer diagnosis is assisted by various techniques and observable symptoms in the human body. An essential aspect is to accurately and precisely diagnose cancer. This work incorporates various ensemble feature selection methods such as Wilcoxon Sign Rank Sum test (WCSRS), Fisher's test (F-test), and Analysis of Variance (ANOVA). The first step of the proposed approach includes data pre-processing, followed by feature selection. The identified gene set is then processed to identify cancer patients through several ensemble learning models, such as Extreme Learning Machines (ELM), Extreme Gradient Boosting (XGBoost), Random Forest (RF) and AdaBoost. The performance of the ELM is enhanced by incorporating the parameters of all ensemble learning models are fine-tuned using the Grey Wolf Optimizer (GWO). The experimentation has been carried out on colon cancer data set, using data from 62 patients, with 2,000 genes analyzed. The feature selection using ANOVA led to the discovery of important features, enabling a Random Forest classifier to achieve an accuracy of 92 %.