This investigation represents an attempt to employ the Devries Suicide Prediction Scale and the Rorschach Inkblot Test in a two-stage predictive model which was designed to decrease the high false positive rate associated with the Devries and to design a way in which the Rorschach could be used efficiently in suicide prediction in a large mental hospital setting. The Rorschach was not found to significantly improve the predictive ability of the Devries. An unexpectedly high percentage of mental patients in the study, thirty-eight percent, admitted to previous suicide attempts, raising the question of whether suicidal behavior is not more common than is usually thought.

This thesis investigates the predictive modeling of suicidal behavioramong older adults in Sweden through the application of machine learning and survival analysis to data from Swedish National Registries. Theresearch utilizes longitudinal and static data characteristics, includingprescribed psychopharmaceuticals, medical conditions, and sociodemographic factors. By incorporating various survival analysis models suchas the Cox Proportional Hazards Model (CoxPH), Random SurvivalForest (RSF), and Gradient Boosting Survival Analysis (GBSA) withsequential machine learning techniques such as Long Short-Term Memory (LSTM) networks, the study aims to enhance the predictive accuracy of suicidal tendencies among the elderly. The thesis concludes thatcombining these techniques provides insights into the risk of suicide attempts and completed suicides while also underscoring the challengesassociated with modeling national registry data and demonstrating anovel approach to addressing suicidal behavior.