Abstract

Suicide is a devastating act in which a person takes their own life. Decades of research into suicide have identified a myriad of risk factors that have been used to create assessments of suicide risk and suicidality. However, more recent research has suggested that these identified risk factors may have no better predictive ability than chance, perhaps because suicide is actually a multi-dimensional, multi-faceted construct that has been viewed too simplistically for prediction’s sake. To try and better appreciate the complex nature of suicide while also increasing prediction accuracy, researchers have turned to machine learning. This study sought to meta-analyze the predictive ability of machine learning in predicting suicide risk. A multi-level, mixed effects meta-analytic model returned a significant model with an effect size of g = 1.36 (p < 0.0001), but with a significant amount of heterogeneity (Q(285 df ) = 66361.51, p < 0.0001). A fully augmented model using three moderators (algorithm type, data source type, and suicide definition) accounted for a significant portion of the variance and also returned a statistically significant model. Meta-regression models showed that algorithm type had a statistically significant effect on the reported effect sizes while data source type and suicide definition did not return significant models. The results of this analysis found not only that machine learning indeed has a significant impact on the accuracy of predicting suicide, but also that the type of algorithm used has a significant impact on the reported accuracies as well. However, high within and between study hetereogeneity warrants more research into other potential moderating variables.

Library of Congress Subject Headings

Suicide--Research; Suicide--Risk factors; Machine learning

Publication Date

8-25-2021

Document Type

Thesis

Student Type

Graduate

Degree Name

Experimental Psychology (MS)

Department, Program, or Center

Psychology (CLA)

Advisor

Esa M. Rantanen

Advisor/Committee Member

Lindsay S. Schenkel

Advisor/Committee Member

Clark Hochgraf

Campus

RIT – Main Campus

Plan Codes

EXPSYC-MS

Share

COinS