Forecast models for suicide: Time-series analysis with data from Italy

Area of Science:

• Public Health
• Epidemiology
• Data Science

Background:

• Suicide prediction is complex, with forecasting models crucial for targeted interventions.
• Existing research often focuses on exploratory analysis rather than predictive analytics for suicide risk.

Purpose of the Study:

• To investigate the accuracy of suicide forecasting models for males and females using time series analysis.
• To compare the performance of forecasting models in predicting monthly suicide numbers.

Main Methods:

• Utilized time series data of 101,499 male and 39,681 female suicides in Italy (1969-2003).
• Split data into training (1969-1996) and testing (1997-2003) sets.
• Assessed model accuracy using Mean Absolute Error, Root Mean Squared Error, Mean Absolute Percentage Error, and Mean Absolute Scaled Error.

Main Results:

• Observed a trend of increasing suicides from 1969, peaking around 1990, then decreasing for both sexes.
• Seasonal and trend components accounted for 24% and 64% of variance in male suicides, respectively; 28% and 41% in female suicides.
• Forecasting models achieved an approximate 10% margin of error, with findings clearer for male suicide data.

Conclusions:

• Forecasting models incorporating seasonality can identify increases (zenith) but struggle with decreases (nadir) in suicide rates.
• Preventative strategies should target factors driving increases above the average trend in seasonal and cyclic suicide patterns.
• The study highlights the utility and limitations of predictive analytics in understanding and mitigating suicide risk.