Truncated spline regression for binary response: a comparative study of nonparametric and semiparametric approaches

Abstract

Type 2 diabetes mellitus is a chronic metabolic disorder with a growing global prevalence, especially in developing countries. According to the International Diabetes Federation, Indonesia ranks fifth in the world for the highest number of diabetes cases, with 19.5 million adults affected by 2021. Early detection and intervention strategies are critical in managing this disease, and predictive models play a vital role in identifying individuals at high risk. Recent advances in regression analysis have introduced nonparametric and semiparametric regression methods, particularly truncated spline-based regression, which offer greater flexibility in capturing complex relationships in data. This study compares the performance of nonparametric and semiparametric truncated spline regression models in classifying binary response variables, specifically in predicting type 2 diabetes mellitus status. The models were evaluated using deviance values and classification accuracy metrics, including sensitivity, specificity, and precision. The results showed that the semiparametric truncated spline regression model outperformed the nonparametric approach, with lower deviance values (42.46 vs 52.94) and higher classification accuracy (86.67% vs 76.67%). In addition, the semiparametric model showed better sensitivity (97.44% vs 92.31%), specificity (66.67% vs 47.62%), and precision (84.44% vs 76.60%), indicating a greater ability to correctly classify diabetic and non-diabetic individuals.