Abstract
Suicide remains a significant public health concern in South Asian countries, with complex underlying factors varying across the region. This study developed and compared machine learning models for predicting age-standardized suicide rates in eight SAARC nations, with the goal of identifying regional patterns and generating future projections. Using WHO Global Health Observatory data from 2000-2019, we implemented three machine learning approaches: Random Forest, Seasonal Autoregressive Integrated Moving Average (SARIMA), and XGBoost. Data preprocessing included handling missing values through Multivariate Imputation by Chained Equations (MICE) and standardizing measurements across variables. Models were evaluated using RMSE, MAE, and MAPE metrics, with country-specific analysis revealing distinct regional patterns in model performance. Random Forest performed best for Afghanistan, Bangladesh, and India (RMSE: 0.245, 0.185, 0.325), suggesting complex non-linear relationships in these larger, more diverse countries. SARIMA showed superior performance for Bhutan, Nepal, and Sri Lanka (RMSE: 0.156, 0.234, 0.345), indicating stronger seasonal patterns in these nations. XGBoost excelled for Maldives and Pakistan (RMSE: 0.128, 0.223), with Maldives showing the most predictable patterns overall. Future projections through 2030 indicate increasing suicide rates across all SAARC countries, with Sri Lanka maintaining the highest rates (24.4 by 2030) and Bangladesh showing the greatest percentage increase (24.2%). These findings highlight the importance of region-specific approaches to suicide prevention and the need for enhanced data collection systems in South Asia.
Presenters
Russell KabirAssociate Professor, School of Allied Health and Social Care, Anglia Ruskin University, Essex, United Kingdom
Details
Presentation Type
Theme
KEYWORDS
SAARC, Machine Learning, Prediction, Suicide
