Performance comparison of logistic regression and random forest model in bipolar disorder diagnosis

Abstract

Abstract. Bipolar disorder (BD) is a significant psychiatric disease that has a large impact on patients living qualities. The diagnosis of BD is important for the early treatment and effective control that benefits the patients, their families, and society. With the development of machine learning algorithms, this paper makes use of the multinomial logistic regression (LR) and Random Forest (RF) models in predicting bipolar disorder (BD). The evaluation and comparison between the two models are conducted to present the possibility of using ML models to help with diagnosis of the BD. A dataset consisting of 120 individuals, including 28 BD Type I patients, 31 BD Type II patients, 31 depression patients, and 30 normal individuals was used in this study. The performance of the models was assessed using metrics such as accuracy, confusion matrix, cross-validation scores, classification report, and ROC-AUC curves. The multinomial LR model demonstrated superior performance with an accuracy of 83%, higher cross-validation scores, and better discriminative ability as indicated by ROC-AUC values. In contrast, the RF model achieved an accuracy of 79%, with lower precision and recall for certain classes. The findings suggest that the multinomial LR model is more effective in predicting bipolar disorder and its subtypes, making it a robust and reliable tool for clinical diagnostics.