Analítica predictiva como apoyo en la salud pública: Modelos de pronóstico con series de tiempo aplicados a la conducta suicida.

Mental health has been identified by the World Health Organization (WHO) as a matter of concern internationally. In Colombia, specifically, the importance of mental health care, including the phenomenon of suicide, has been highlighted. In this regard, policymakers have sought alternatives to address or mitigate this issue. In this context, research becomes relevant regarding the development of tools that facilitate decision-making for governmental authorities, for example, through the formulation of forecasting models that enable the identification of trends and patterns of behaviour of suicide attempts. The objective of this study is to formulate and select a regional forecasting model for suicide attempts by means of predictive analytics techniques. This paper contributes to fill the gap in the case of suicide attempts in the city of Medellín, Colombia. The fitting, validation and comparison of three different time series models, under minimum forecast error criteria is presented. The compared models include parametric, Holt-Winters and Box-Jenkins approximations; For the analysed data, the parametric model with cubic, seasonal and ARMA(0.5) components is the one with the lowest forecast error. This model manages to capture the trends of the phenomenon, with a low level of error for the projection for nearby trends, but it does not manage to respond to sudden changes in structure such as those that occurred in the COVID-19 pandemic. [ABSTRACT FROM AUTHOR]