Identifying Influential Predictors of Loneliness in a Thai Community: A Cross-Sectional Machine Learning Analysis

Abstract

This research addresses the increasing public health concern of loneliness within the Thai community by  developing and validating a Machine Learning (ML) model for risk detection. The objective is to use ML to identify the most influential psychological, physical, and socioeconomic factors contributing to loneliness, providing insights that extend beyond standard self-assessments. A cross-sectional, multi-platform data collection approach was utilized, employing the LINE application (the "Senior Thailand" Official Account) and the "senior.in.th" website to gather comprehensive information from 616 adults across various provinces in Thailand. The methodology included all 12 factors for predictive modeling, encompassing sociodemographic traits, lifestyle habits (e.g., exercise and Activities of Daily Living), physiological metrics (e.g., BMI and blood pressure), and detailed mental health assessments (e.g., UCLA Loneliness Scale and Depression/Anxiety scales). Four algorithms Decision Tree, Random Forest, Logistic Regression, and Support Vector Machine (SVM) were trained and evaluated. The Logistic Regression model showed the best performance, achieving a Classification Accuracy (CA) of 0.707 and the highest Area Under the Curve (AUC) of 0.697. However, the low Matthews Correlation Coefficient (MCC) of 0.253 and the moderate AUC suggest that, while the model has utility, the complexity of loneliness requires further hyperparameter tuning and model optimization. Overall, this study successfully validates a multi-factor ML model for loneliness prediction, revealing that the most significant predictors are psychological and functional barriers (such as depression and Activities of Daily Living). This outcome highlights the potential for ML to inform targeted public health strategies and promote social well-being by prioritizing interventions based on data-driven insights.