Development of a Web-Based System for Predicting Depression and Suicide Attempt Using Ensemble Machine Learning Model

Abstract

Depression is a mental illness that can make a person’s life difficult and can eventually lead to suicide. Depressed individuals who do not receive timely attention develop worse conditions and may eventually commit suicide. Depression and suicide are becoming a global health concern which need to be adequately addressed. In this study, an ensemble learning model which make use of demographic data to detect depression and suicide attempt and also guide individuals from committing suicide through the web-based application system is proposed. The forever Alone demographic dataset which was downloaded from Kaggle online data repository was used, the dataset was imbalanced and was balanced using synthetic minority oversampling technique (SMOTE). The dataset was split into 60/40, 70/30 and 80/20 train/test percentage split, however, the 80/20 train/test split performed best and it was used and reported in this study. The study employs an ensemble machine learning model, specifically Adaboost with Extra trees as base estimators for prediction. Adaboost enhances model performance especially in handling class imbalance leading to excellent accuracy.  Results obtained reveal that Adaboost ensemble model outperformed all other machine learning algorithms across all evaluation metrics with 82.00% recall and 78.69% accuracy for depression, and 93.85% recall and 90.60% accuracy for suicide attempt respectively on the balanced dataset.  The uniqueness of Adaboost in sequential weighting of misclassified instances which enhances model performance, especially in handling class imbalance thus leading to an excellent accuracy. It was therefore used for the prediction system. The study affirmed the prowess of ensemble machine learning model for predicting depression and suicide attempt. Ethical issues were also discussed in the study.