A Hybrid Ensemble Approach for Depression Detection: Combining Deep Learning and Machine Learning

The depression detection from text data represents a pragmatic application of machine learning and natural language processing in the mental health field. This paper presents a hybrid ensemble learning approach that combines conventional machine learning models, namely Naive Bayes, Random Forest, XGBoost, and Logistic Regression, with deep learning models, including the Long Short-Term Memory (LSTM) and Multi-Layer Perceptron networks. The hard voting for ensembling results in the ensemble model outperforming individual models on precision, on a par with the best model in terms of accuracy, F1-score, and recall. The evaluation of the models on a dataset of depression-related text posts indicates that the ensemble significantly reduces both false positives and false negatives, shown in the confusion matrix. The LSTM model, integrated into the ensemble, captures sequential dependencies in text, contributing to improved depression classification. This study demonstrates the interplay potential of machine learning and deep learning techniques to improve the balance and robustness of depression detection systems.