Forecasting Noncommunicable Diseases in Thailand: Evaluating the Predictive Power of Social Determinants of Health-Related Features

Abstract

In Thailand, non-communicable diseases (NCDs) presented a significant health and economic challenge. This study investigated machine learning (ML) for predicting NCDs prevalence using social determinants of health (SDHs). Two scenarios, baseline and inference (imputing missing values) were assessed. Monthly household expenditure and hospital counts appeared as pivotal features in the inference scenario. Model performance remained comparable between scenarios, with slight variations for specific NCDs. Random Forest (RF) showed slightly superior predictive power (RMSE: 1.53 -- 74.93, R Square: -0.11 -- 0.11), though interpretability remains a challenge. Addressing data limitations and enhancing interpretability are crucial for fully harnessing ML's potential in NCDs prediction and prevention. The study's findings underscore the importance of integrating ML and SDHs into public health policy to effectively combat NCDs in Thailand, potentially saving lives and fostering sustainable socio-economic development. The study revealed the intricate interplay between socio-economic factors and NCDs prevalence, emphasizing the need for targeted interventions. The exploration of machine learning algorithms in predicting NCDs prevalence provided valuable insights into model performance and highlighted the significance of features such as household expenditure and hospital counts. Moving forward, efforts to address data disparities and enhance model interpretability are essential to maximize the utility of predictive modeling in informing public health policies aimed at mitigating the impact of NCDs in Thailand.