Assessing Feature Selection Techniques for a Colorectal Cancer Prediction Model

Abstract

Risk prediction models for colorectal cancer play an important role to identify people at higher risk of developing this disease as well as the risk factors associated with it. Feature selection techniques help to improve the prediction model performance and to gain insight in the data itself. The assessment of the stability of feature selection/ranking algorithms becomes an important issue when the aim is to analyze the most relevant features. This work assesses several feature ranking algorithms in terms of performance and robustness for a set of risk prediction models. Experimental results demonstrate that stability and model performance should be studied jointly as RF turned out to be the most stable algorithm but outperformed by others in terms of model performance while SVM-wrapper and the Pearson correlation coefficient are moderately stable while achieving good model performance.