Natural Language Processing Algorithms to detect mental disorders in College Students in Lima, Perú

This study conducts a comparative analysis of six Natural Language Processing (NLP) models for detecting mental disorders among university students: BERT, Support Vector Machine (SVM), Random Forest, FLAN-T5, Mental-Llama, and a Customized NLP Model (C-NLP). We utilized a dataset of 10,000 textual responses from college students in Lima and texts sourced from platforms like Reddit, which encompass a diverse array of emotions and mental states. The models were evaluated using statistical significance tests such as p-values and confidence intervals alongside traditional metrics including precision, accuracy, F1-score, and AUC-ROC. The results indicate that both BERT and the C-NLP achieved the highest accuracy of 92%. The custom model effectively adapted to Peruvian students' cultural and linguistic nuances, providing a balanced approach between performance and cultural relevance.