A MULTIMODAL APPROACH TO MENTAL HEALTH PREDICTION: MACHINE LEARNING PERSPECTIVE ON LIFESTYLE AND BEHAVIOURS

Abstract

Depression in particular, is now a major and increasing concern for students everywhere. These problems are usually made worse by students having to deal with pressures from school, society, and changes in their life, all of which leave them more at risk. The use of self-report surveys and periodic counseling tends to make it challenging to notice depressive symptoms that are not easy to detect. To deal with the drawback explained above, the current analysis presents an advanced system that relies on different machine learning (ML) models. Random Forest, XGBoost, Logistic Regression, LightGBM, Support Vector Classifier (SVC), and K-Nearest Neighbors (KNN) are some of the models used in this system to detect depression in students. To build the predictive models, we used large numbers of distinct features coming from demographics, academic results, and daily routines. To tackle the class balance problem, SMOTE was used, making sure the data in the two classes was more even. Carrying out many preprocessing tasks helped the models to perform higher quality work. To do this, missing values were taken care of through appropriate imputation, duplicate or inconsistent data was removed, filtering based on values was used to keep the important parts, and statistical methods were used on outliers to reduce errors in the dataset. Evaluating how accurate the model performed showed that both XGBoost and LightGBM were 97% accurate. The study proves that machine learning algorithms can help identify early mental health problems, even when the symptoms are not very noticeable. The results supports the use of AI-based approaches to early identify mental health risks and offer the relevant guidance for supported interventions to help students that are affected by depression. More studies will attempt to connect digital behavior information with AI algorithms to make the predictions more accurate. Moreover, XAI elements will be used to make the model transparent and instill trust in its users. Testing the system among students from different cultures and countries is essential to find out if it can be used by people the world over.