Design and Simulation Based Evaluation of Artificial Intelligence Driven Personalized Mental Healthcare Support System Integrating Natural Language Processing and Machine Learning

Abstract

Mental-health services face limited accessibility, delayed assessment, little personalization and a shortage of specialists. This paper proposes the Personalized Mental Healthcare Support System (PMHSS), a decision-support framework that integrates natural language processing (NLP), emotion/sentiment analysis, machine learning (ML), a personalized recommendation engine and a conversational AI assistant. PMHSS registers users securely, administers standardized questionnaires, analyses free-text messages, fuses questionnaire, emotional, profile and historical features to classify each user into a Low, Medium or High mental-health risk class, and then recommends self-care, CBT-oriented or professional-support resources. Further modules provide chatbot interaction, counsellor recommendation, appointment scheduling, encrypted teleconsultation and longitudinal progress monitoring. Because no real patient data were used, the risk-classification module was evaluated on a synthetic cohort of 10,000 simulated users (7,000 training, 1,500 validation and 1,500 test users; random seed 42). On this cohort PMHSS reached an accuracy of 98.2%, precision of 98.0%, recall of 97.3% and F1-score of 97.6%, compared with 92.4%, 91.8%, 90.9% and 91.3% for logistic regression, 95.7%, 95.1%, 94.3% and 94.7% for random forest, and 97.1%, 96.7%, 96.0% and 96.3% for XGBoost. These figures describe performance on simulated data only and do not demonstrate clinical validity. Emotion recognition, recommendation quality, chatbot safety and the scheduling and teleconsultation modules were not evaluated. Validation on real, clinician-labelled data, together with safety and fairness audits, is required before any clinical use.