LASSO and Elastic Net in the Prediction of Dementia

Abstract

Dementia, referred to in medical terms as 'major neurocognitive disorder,' is a prevalent condition impacting a substantial number of individuals in the United States, where approximately 5.5 million people currently grapple with the ailment. Despite its widespread occurrence and the profound consequences, it inflicts on individuals and their families, dementia remains underdiagnosed and poorly recognized [1]. This study aim to investigate the correlation between demographic factors and dementia, employing cross-tabulation and the chi-square test. Additionally, it aimed to compare the predictive capabilities of LASSO and Elastic Net regression models in forecasting dementia occurrences using AICc and BIC. Finally, the third objective was to identify the primary factor influencing dementia utilizing the most effective model. The data set implemented in this study is longitudinal dementia data obtained from MIT Press Direct web. The study's results revealed that logistic LASSO regression demonstrated superiority in predicting dementia. Furthermore, the research delineated that significant factors influencing dementia encompass gender, age, years of education, and socioeconomic status. These findings provide valuable insights for healthcare authorities and governments in both the United States and Malaysia, enabling them to allocate adequate medical resources and formulate guidelines tailored to the ageing population in specific districts of the country.