Machine Learning and Deep Learning Models for Predicting Mental Health Disorders and Performance Analysis through Chatbot Interactions

Mental health disorders have recently been prompting increased concern globally and finding new ways of diagnosing and treating them efficiently. Machine learning (ML) and deep learning (DL) enabled chatbots are enormous tool for predicting and supporting mental health. This work aims to carry out an assessment of several AI models for prognostics of mental health disorders based on the comparison of intents, patterns, and responses in a structured chatbot-based dataset. Since it is intent-based, our dataset is best suited to classifying user inputs accurately into different mental health thematic buckets such as anxiety, stress, and proved suicide ideation. To assess the models, we compared basic models such as Multinomial Naïve Bayes, Random Forest and SVM as well as deep learning models including LSTM networks. SVM and LSTM showed promising results among the tested models with the accuracy of 94.6%. LSTM was proved to address the problem of sequential context dependence typical for conversational data. For further improvement in the model’s accuracy, we used ensemble methods whose accuracy came out near like the highest accuracy models, 94.2% accurate. This work is new in the sense that it involves the use of data from an intent-based chatbot, and a comparison of the ML and DL models designed specifically for the prediction of mental health outcomes. Also, it is important to note that we dealt with underrepresented intents, including suicide ideation, using data augmentation and ensemble approach. It fills the gaps in the deployment of AI for mental health by providing recommendations concerning the model’s performance and possible ethical concerns as well as integrating it into conversational assistance. We also found the relevance of an AI chatbot in the delivery of efficient and easily deployable intervention for mental health.