A Comparison of Linear Mixed Effect Models and REEM Trees for Prediction of Cognitive Decline
Kristina Vatcheva
University of Texas Rio Grande Valley
Cognitive decline is common with ageing, but other risk factors may influence this process. Cognitive decline can have profound implications for individuals' well-being and its prediction and early detection can prevent and improve lives and decrease hospitalization cost. We compared the performance of linear mixed effects model and RE-EM tree on predicting cognitive decline. Data from five waves of the English Longitudinal Study of Aging (ELSA) were analyzed. RE-EM trees using 1 and 6 iterations and three linear mixed effects models, with predictors selected by RE-EM trees and with all predictors, with random intercept and a slope for time variable, were fitted on training data. Models' prediction abilities were evaluated on test data using root mean squared error (RMSE). Data were unbalanced and comprised of 12, 212 participants with a total of 42, 560 records. All liner mixed effects models resulted with better prediction performance compared to the fitted RE-EM trees (RMSE=3.57, RMSE=3.60, RMSE=3.63 vs. RMSE=3.67 and RMSE=3.68, respectively).