HyOPTEnsemble: custom-weighted soft voting hyperparameter optimization ensemble model, explainable-AI for predicting mental state among university students

The emergence of mental health issues among university students has a detrimental impact on their academic and job performance, leading to increased levels of anxiety, dropout mentality, and even suicide. Male students are at higher risk compared to their female counterparts. Early detection of mental health issues is crucial for effective treatment. While previous research has focused primarily on specific departments or universities, few studies have examined data from multiple institutions. This study proposes a novel hypertuned ensemble machine learning model that demonstrates its effectiveness in predicting the mental state of university students based on their past two weeks’ experiences. The dataset was collected through an online survey conducted over a period of one month, involving 50 universities, with a total of 400 responses from undergraduate and graduate students. Of these, 67.1% were male, 31.7% were female, and 1.2% declined to disclose their gender identities. To address the data imbalance issue, a hybrid synthetic minority oversampling method, SMOTE-ENN, is employed. Initially, single machine learning models were utilized, including Decision Tree (DT), Naive Bayes (NB), Ada-Boost, Bagging, Random Forest (RF), and Modified Random Forest. In addition, a grid search was performed for hyperparameter optimization. However, the ensemble model with 5-fold cross-validation achieved the highest accuracy of 0.977, along with outstanding performance metrics of precision, recall, and f1 score, at 0.983, 0.977, and 0.979, respectively. Furthermore, Cohen’s kappa coefficient and Matthews correlation coefficient were utilized to measure reliability. The proposed model demonstrates the ability to predict the mental state of university students.