A Classification Based Machine Learning Model to Predict Suicidal Thoughts

Abstract

Suicide is an unnatural death and it becomes a major issue in Bangladesh. The World health organization report which is published in 2018, it says, Suicide Deaths in Bangladesh come across 9,544 deaths, or in percentage, it is 1.23% of entire deaths [1]. We aimed to develop a model that can predict suicidal thoughts, using a machine learning algorithm. It can prevent future risk of suicidal attempts. Dataset presents 15-46 years old people's thoughts, feelings, their regular activities, contains a total of 22 attributes and 441 instances. The classification process was performed using nine machine learning algorithms those are, Naive Bayes, KNN, Linear SVC (support vector classifier), Non-linear SVC, Random Forest Classifier (RFC), Decision Tree, Logistic Regression (LR), and Extreme Gradient Boosting (XGB) Classifier, Adaptive Boosting(Ada-boost) Classifier. The prediction model achieved a good performance. The highest accuracy achieved Random Forest Classifier (0.91). The area under the receiver operating characteristic curve (AUC)=0.9 for Random Forest Classifier. This study shows the probability that a machine learning approach can able to decrease suicide risk. Hopefully, this model will assist as a support for reducing future suicidal risk. The paper ends with a review of various practical issues, which may be explored to enhance model performance.