You are here

Classification of Suicide Ideators and Attemtpers with Machine Learning Techniques

Title: Classification of Suicide Ideators and Attemtpers with Machine Learning Techniques.
Name(s): Huang, Xieyining, author
Franklin, Joseph, professor directing thesis
Joiner, Thomas, Jr., committee member
Wagner, Richard K., committee member
Florida State University, degree granting institution
College of Arts and Sciences, degree granting college
Department of Psychology, degree granting department
Type of Resource: text
Genre: Text
Master Thesis
Issuance: monographic
Date Issued: 2018
Publisher: Florida State University
Place of Publication: Tallahassee, Florida
Physical Form: computer
online resource
Extent: 1 online resource (43 pages)
Language(s): English
Abstract/Description: Background: Suicide is a major public health concern. To facilitate treatment and prevention efforts, prior research has focused on the identification of a small set of factors that might differentiate between ideators and attempters. Recent findings from meta-analyses and machine learning studies, however, indirectly suggest that the differences between ideators and attempters might be much more complex than previously theorized. This study aims to directly test whether the differences are simple or complex by adopting both traditional statistical methods and machine learning approaches. Method: A total of 285 participants who have either thought about suicide (N = 131) or attempted suicide (N = 154) in their lifetime were recruited. Participants completed questionnaires examining risk factors for suicidal thoughts and behaviors. Statistical models ranging from simple to complex were adopted to classify ideators and attempters (i.e., univariate and multivariate logistic regressions, random forests with cross-validation and bootstrap optimism to adjust for overfitting). Unsupervised machine learning (i.e., K-means clustering) was used to identify two underlying subgroups among the sample. Results: Overall, more complex algorithms were more adept at distinguishing between ideators and attempters. Univariate logistic regressions on average produced poor classification accuracy (AUC = 0.54); multivariate logistic regression yielded similar results (AUC = 0.62). Random forests with cross-validation to safeguard against overfitting classified ideators and attempters with fair accuracy (AUC = .0.78). Random forests with bootstrap optimism produced good accuracy (AUC = 0.89). Unsupervised machine learning appeared to identify two subgroups based on psychopathology severity, a grouping inconsistent with ideator/attempter status (AUC = 0.59). Discussion: Consistent with previous research, this study showed that the differences between ideators and attempters are likely complex. Future studies are encouraged to replicate and extend the current study by adopting different study designs with larger sample sizes and variables.
Identifier: 2018_Sp_Huang_fsu_0071N_14339 (IID)
Submitted Note: A Thesis submitted to the Department of Psychology in partial fulfillment of the requirements for the degree of Master of Science.
Degree Awarded: Spring Semester 2018.
Date of Defense: March 5, 2018.
Bibliography Note: Includes bibliographical references.
Advisory Committee: Joseph C. Franklin, Professor Directing Thesis; Thomas E. Joiner, Committee Member; Richard K. Wagner, Committee Member.
Subject(s): Clinical psychology
Persistent Link to This Record:
Host Institution: FSU

Choose the citation style.
Huang, X. (2018). Classification of Suicide Ideators and Attemtpers with Machine Learning Techniques. Retrieved from