Classification and detection of mental health disorders in university students using a machine learning approach based on text datasets
Keywords:
Early Detection, Machine learning, Mental Health, Text `Classification, StudentsAbstract
Student mental health is an increasingly critical issue in the academic environment, yet early detection remains very limited. This study aims to automatically classify student mental health conditions using
machine learning techniques based on text analysis and survey data. The datasets used consist of five sources: (1) Student Mental Health Survey (101 respondents), (2) Combined Mental Health Dataset
(53,043 statements with 7 mental condition classes), (3) Mental Health Text Dataset (27,977 binary classification text samples), (4) Mental Health FAQ Dataset (98 question-answer pairs), and (5) Chatbot
Training Dataset (3,512 context-response pairs). The results of the survey analysis indicate that 34.6% of students experience depression and 33.7% experience anxiety, but only 5.9% seek professional help—
indicating a worrying treatment gap. Classification experiments using the Gradient Boosting algorithm achieved the best performance with 89.2% accuracy and an F1-score of 89.1% on the binary classification task, and a macro F1-score of 81.4% on the multiclass classification task. The application of the SMOTE technique consistently improved performance on the minority class by an average of 8.3 percentage points. This research contributes to the development of an artificial intelligence-based early detection framework to support mental health interventions in Indonesian campus environments.
