Stroke is a non-communicable disease that affects the brain and can lead to motor, sensory, and cognitive disabilities due to damage such as cerebral small vessel disease. The stroke, which consists of ischemic and hemorrhagic types, is one of the leading causes of death and disability worldwide. In this study, we model the risk of both ischemic and hemorrhagic stroke using a nonparametric binary logistic regression approach based on the least squares spline estimator. This model flexibly estimates odds ratios, enabling it to capture nonlinear relationships between risk factors and stroke occurrence. The analysis was conducted using secondary data from Dr. Drs. M. Hatta Brain Hospital in 2023, with risk factors including LDL, uric acid, triglycerides, blood glucose, sodium, and age. The results show that for age more than and equal to 50 years, both LDL and triglyceride levels more than and equal to 125 mg/dL, and blood glucose levels across all spline segments are associated with increased odds of ischemic stroke. Meanwhile, uric acid and sodium levels across all spline segments show a decreased tendency toward ischemic stroke risk. The model achieved an accuracy and AUC of 87.14% and 0.892, respectively, for in-sample data, and 90% and 0.911 for out-sample data. These findings demonstrate that spline-based nonparametric binary logistic regression provides a more flexible and accurate approach for modeling stroke risk. This study also supports the achievement of the SDGs by contributing to data-driven early detection and stroke prevention strategies.

This study aims to develop and apply a Mixed Estimator Nonparametric Regression–Spline Truncated and Fourier Series (MENR–SF) to model the nonlinear relationships between behavioral factors and the prevalence of heart disease in Indonesia. The proposed approach simultaneously combines spline truncated estimators and Fourier series within a unified nonparametric regression framework, allowing each predictor variable to be modeled according to the specific characteristics of its relationship with the response variable. Parameter estimation is conducted using the Least Squares method, while the optimal number of spline knots and Fourier oscillations is determined based on the Generalized Cross-Validation (GCV) criterion. The application of the MENR–SF model to data from the 2023 Indonesian Health Survey (Survei Kesehatan Indonesia, SKI), with 38 provinces as the units of analysis, indicates that the best-performing model is obtained when the prevalence of daily smoking, the proportion of insufficient physical activity, and habitual consumption of fatty foods are modeled using spline truncateds, whereas the proportion of hypertension control is modeled using a Fourier series. The optimal combination, with three spline knots and three Fourier oscillations, yields a minimum GCV value of 1.197, low prediction error, and a coefficient of determination of 0.94, indicating an excellent ability of the model to explain variations in heart disease prevalence. These findings conclude that MENR–SF is a flexible and accurate approach for modeling complex nonlinear relationships in health data. The model offers enhanced flexibility and richer interpretability regarding the effects of behavioral factors, thereby holding strong potential to support data-driven health analysis and policy formulation.