Using Machine Learning Method in Assessing the Significance of Socio-Psychological and Molecular- Genetic Factors Associated with Suicidal Behavior

Abstract

Introduction. The use of machine learning can significantly improve the prediction of the probability of committing suicide, compared with methods based on classical statistical approaches. Purpose. To study the socio-psychological and molecular genetic factors of suicidal risk using machine learning methods. Materials and methods. The study involved 299 people aged 18 to 75 years. The study group consisted of people who committed self-harm in various ways, people who were in a state of adjustment disorder and did not make suicidal attempts, and young people aged 18-27 who had no mental disorders at had not previously committed suicide attempts. In the course of the analysis, social status (education, income, employment, marital status, upbringing characteristics), individual characteristics (G. Eysenck’s personality questionnaire; G. Schmishek – K. Leonhard’s test questionnaire) and the frequency of occurrence of genotypes and alleles of the HTR1 genes were assessed BDNF, COMT, SKA 2. When evaluating the obtained data, the following algorithms were used: SVC ROC; RandonForest ROC; KNeighborsClassifier ROC; LogisticRegression ROC from Python programming library. Results and discussion. Statistical significance was obtained for 4 predictors: the presence of higher education, the absence of punishment in childhood, the presence of the HTR1A G/G genotype, and the level of demonstrativeness. Conclusions. 1. The presence of higher education and the absence of punishment in childhood is a factor in the prediction of suicides. 2. It is important to have not only a genetic predisposition, but also character traits, causing a person to be in a state of chronic stress. 3. The use of machine learning makes it possible to critically evaluate established ideas about suicidal risk factors and develop new approaches to the prevention of suicidal behavior.