A novel two-tier feature selection model for Alzheimer’s disease prediction

Abstract

The interdisciplinary research studies of artificial intelligence in health sector is bringing drastic life saving changes in the healthcare domain. One such aspect is the early disease prediction using machine learning and regression algorithms. The purpose of this research is to improve the prediction accuracy of Alzheimer ’s disease by analysing the correlation of unexplored Alzheimer causing diseases. The work proposes Chi square-lasso ridge linear (Chi-LRL) model, a new two-tier feature ranking model which recognizes the significance of including diabetes, blood pressure and body mass index as potential Alzhiemer predictive parameters. The newly added predictive parameters of Alzheimer’s disease were statistically verified along with the conventional prediction parameters using chi-square method (Chi) as Tier 1 and an embedded model of lasso, ridge and linear (LRL) Regression for feature ranking as Tier 2. The performance of the proposed Chi-LRL model with selected features were then analysed using machine learning algorithms for performance analysis. The result shows a noticeable performance by selecting eleven significant features and a 4.5% increase in the prediction accuracy of Alzheirmer disease.