The relationship between mental health and academic performance is a vital area of research, particularly among university students. In this paper, we present an integrated multimodal and hybrid framework for mental health prediction, utilizing both textual and visual data to improve detection accuracy. This study explores the effects of mental health on students’ CGPA using a dataset collected through a survey conducted via Google Forms, publicly available on Kaggle. The dataset comprises responses related to students’ mental health and academic performance. Preprocessing techniques, including tokenization, lemmatization, and embedding like BERT, was applied to extract meaningful patterns from the textual data. A hybrid ensemble framework, integrating Naive Bayes, stacking, and XGBoost, was implemented to enhance predictive accuracy and robustness. The proposed system achieved an accuracy of 85% with Naive Bayes and 80% with Random Forest, validated through metrics such as precision, recall, and F1-score. The results highlight significant correlations between mental health factors and academic outcomes, emphasizing the effectiveness of machine learning in identifying these relationships. This research provides a scalable framework for analyzing mental health influences on academic success and sets the stage for future studies incorporating multimodal datasets.