Space-time clustering and Bayesian network modelling of suicide dynamics in India

Suicide represents a major global public health issue, necessitating a detailed analysis of its spatiotemporal patterns and the complex interplay among various contributing factors. This study employs Bayesian network classification and modeling techniques, integrating spatiotemporal dimensions to investigate suicide dynamics comprehensively. We analyze the spatiotemporal distribution of attempted (ATD) and completed (CTD) suicide events across Indian states from 2017 to 2022, utilizing data from government-authorized sources such as the National Crime Records Bureau (NCRB) and the Census. The SaTScan software facilitates space-time analysis, while Python is used for variable classification and modeling. Performance evaluation reveals that the Tree Augmented Naive Bayes (TAN) model surpasses baseline, ensemble, and Bayesian models, achieving accuracies of 73% for ATDs and 64% for CTDs. Our findings identify key explanatory variables influencing both types of suicide events based on their causal probabilities. This Bayesian network modeling underscores the interdependence of factors, demonstrating that the impact of each variable is intricately linked to others. Overall, this research establishes a comprehensive framework for mapping the spatiotemporal distribution of suicide attempts and completions, elucidating the complex relationships among associated factors. These insights are crucial for informed decision-making and the development of targeted interventions.