A Comparison of Decision Tree, Logistic Regression, Artificial Neural Network and Random Forest Algorithms to Predict Suicidal Ideation Among Young Adults in Malaysia

Abstract

Suicide is a significant global public health issue, and Malaysia is no exception, with a high incidence rate. On average, approximately 10 suicide deaths occur daily in the country, alongside numerous attempted suicides. Hence, the key indicators for suicidal ideation should be identified so that communities can be aware of the characteristics of suicide attempters and assist them. Therefore, this study aims to develop predictive models using four predictive techniques, which are Decision Tree, Logistic Regression, Artificial Neural Network and Random Forest to anticipate suicidal ideation among young adults in Malaysia. By analysing key indicators, such as demographic, socio-economic, and psychological factors, the model seeks to enable proactive intervention and support for vulnerable individuals. A total of 33 predictive models are generated and evaluated based on their performance using the misclassification rate. Among these models, Gini decision tree models with 2 and 3 branches (80:20) showed superior performance, with the lowest misclassification rate recorded at 19.44%. Consequently, the model with 2 branches is selected for its practicality and accuracy in identifying vulnerable individuals. Early intervention is crucial in identifying and supporting young adults at risk of suicidal ideation. The developed predictive model offers valuable insights for proactive intervention and support, aiding in prevention efforts and reducing the prevalence of suicide. It carries significant policy implications for suicide prevention in Malaysia, enabling targeted intervention strategies for vulnerable young adults. By prioritising resources based on the identified risk factors, policymakers can enhance mental health support systems and prevent tragic outcomes.