Suicide Rate Forecasting
The Problem
Suicide is the 12th leading cause of death in the United States, claiming over 49,000 lives annually. Mental health organizations and governments allocate crisis resources — hotlines, counselors, treatment centers — but often reactively, after trends have already worsened.
The question I asked: Can time-series forecasting provide reliable multi-year projections to help public health agencies plan proactively instead of reactively?
The Approach
- Data Collection: Combined WHO global suicide statistics with CDC national data, covering multiple decades and demographic breakdowns
- Time-Series Decomposition: Separated trend, seasonality, and residual components using STL decomposition to understand underlying patterns
- ARIMA Modeling: Auto-ARIMA for optimal (p,d,q) parameter selection; validated stationarity with ADF test and residual diagnostics
- Prophet Modeling: Facebook Prophet for capturing multiple seasonality patterns and holiday effects with uncertainty quantification
- Demographic Analysis: Separate forecasts by age group, gender, and region to identify at-risk populations
Key Results
- R² = 0.85 on held-out test period, demonstrating reliable forecasting accuracy
- Prophet outperformed ARIMA on datasets with strong seasonality and structural breaks
- ARIMA provided tighter confidence intervals for short-horizon forecasts
- Demographic breakdowns revealed disparate trends across age groups, enabling targeted intervention planning
- Uncertainty intervals help planners prepare for best-case and worst-case resource needs
Business Value
Resource Allocation: Accurate forecasts help health departments budget for crisis counselors, hotline staff, and treatment beds years in advance rather than scrambling reactively. Lives Saved: Proactive resource placement in at-risk regions can reduce response time from days to hours. Technical Transfer: The same ARIMA/Prophet framework I used here is exactly what I deployed at Accenture for demand forecasting across 60K+ SKUs.