Abstract
<p><span lang="EN-SG">Diabetes Mellitus (DM) is a high-risk metabolic disease with increasing prevalence in Indonesia, requiring an effective classification model based on significant risk factors. This study uses Nonparametric Ordinal Logistic Regression based on the Multivariate Adaptive Regression Spline estimator (NOLR-MARS). Unlike conventional parametric ordinal regression, this model does not assume a fixed functional pattern but rather determines the form of the relationship based on data patterns through basis functions, making it more flexible in handling complex predictor variable interactions. Using 664 records from the Non-Alcoholic Fatty Liver Disease (NAFLD) cohort, we explore the relationship between metabolic factors, included age, sex, Body Mass Index (BMI), LDL cholesterol, and hypertension—and DM risk. This NOLR-MARS integration addresses the nonlinear relationship while maintaining the ordinal nature of DM stages, a combination often overlooked in traditional models. Based on Generalized Cross Validation (GCV) selection, the best model achieved 74.92% accuracy for in-sample data and 80.30% for out-sample data. Furthermore, a sensitivity of 70% and a specificity of 92.86% were obtained for stage 2 DM. Factors such as age, BMI, LDL cholesterol, and hypertension significantly influenced DM status. The results showed that the NORL-MARS model had good predictive performance. The novelty of this study lies in the integration of the MARS estimator into an ordinal logistic regression framework for more granular DM risk assessment. Although this model shows potential as a screening tool in high-risk metabolic cohorts, further clinical application requires external validation to ensure broader generalizability.</span></p>