Statistical Models and Machine Learning in Depression Detection: Evaluating XGBoost
TL;DR
The research compares the performance of XGBoost with traditional logistic regression and decision tree approaches for identifying and predicting depression and argues that XGBoost has the potential to improve current diagnostic practices by reducing current time and costs.
Abstract
The research compares the performance of XGBoost with traditional logistic regression and decision tree approaches for identifying and predicting depression. It utilizes the Behavioral Risk Factor Surveillance System, which includes questions about mental and physical health, as well as some sociodemographic variables. XGBoost achieved 75.82% accuracy with an F1 score of 0.5385 and an AUC-ROC of 0.7649, demonstrating superior performance in the depression classification task, particularly with complex and dissociated data structures. Furthermore, the research highlights the importance of timely depression recognition for effective treatment and argues that XGBoost has the potential to improve current diagnostic practices by reducing current time and costs. The research also addresses the necessary future provisions for this approach and its ethical implications in healthcare.