Forecasting Suicides in US for Allocating Counsellors

Project Details




Aniket Singh, Mithun Mohandas, Rahul Agrawal, Saurav Basu, Vijay Swaminathan





This report focusses on creating monthly forecasts of suicides using firearms for the year 2015.
Action Alliance is a big organization with operations in various social areas. With such large
requirements of contractual work force, there is huge scope of cost savings by efficient human
resource allocation.
This forecasting exercise predicts with 95% accuracy, monthly suicides involving firearms. The
model also predicts the deaths by gender, age and location of death. With a year’s view of the
deaths statistics, Action alliance will be able to point out months of low and high deaths thereby
pre-preparing for contingencies. Efficient allocation also improves effectiveness of counselling
as the right volunteers can be procured by observing categorized series (gender, age, location of

The table shows the %increase in number of suicides compared to the same month in the
last year. There is a marked increase in the months of February and May for the total number of
suicides. Similarly, there is a marked increase of 14% in the number of suicides committed by
youngsters under the age of 20. There is a 22% forecasted increase in the number of suicides in
May within the same age group. Similar trends can be observed by location of the incident.

Data Source and Description – Overall firearm deaths data (monthly data from 2012 to 2014)
was sourced from Kaggle ( for this
analysis containing over 100K deaths. Data was filtered for suicides (up to 66% cases are
suicides, rest being homicides and accidents). There are an average of 1800 monthly suicides,
~1000 homicide incidents and ~50 accidents involving firearms. Forecasts predict an average
increase of 4% in suicides in the next year. Charts below show the overall gun deaths split by
suicides, homicides and accidents. Graphs show yearly seasonality, no trend and noise in few
months. It has details of each individual who died in this period – age of the person, race, place
of death, educational qualification and gender. It comprised of suicide, homicide and accidental
cases of deaths by firearms.

Forecasting Methods - Forecasts were generated using naïve, smoothing and multiple linear
regression models. Final forecasts were created after comparing a host of methods (details
below) and choosing data points from most favorable methods. Various models were trained
using data for 24 months and then validated using another 12 months. Same models were then
used to create another 12 months of forecasts. Upon comparing all methods, Holt’s Winter
Additive method was found to be the most suitable candidate for forecasting.

Conclusion and Recommendation
Monthly suicide numbers should be used for planning and setting up contracts for sourcing of
contract employees a year ahead of time saving considerable costs. Unusual busy periods could
be identified for better readiness. Though the model forecasts for 12 months into the future, this
exercise should be repeated every 6 months to include actual data and re-create future forecasts
for planning considering the fact that the socio-economic and political space in undergoing rapid
change in the US at present.

Application Area: