Psychological disorders such as stress, depression, and anxiety remain highly prevalent and exert considerable negative effects on personal well-being. Effective recognition and analysis of these conditions are vital for minimizing their detrimental consequences. This investigation applies machine learning methodologies to advance mental health evaluation. The developed “Stacking-Classifier-Ensemble-Learning” methodology aggregates multiple weak learners to strengthen overall predictive capacity. Base learners comprised neural networks, decision trees, random forests, Naive Bayes, support vector machines (SVM), K-nearest neighbors (KNN), linear discriminant analysis, and AdaBoost, with SVM functioning as the meta-learner. Class imbalance prevalent in mental health data was mitigated through the Synthetic Minority Over-sampling Technique (SMOTE), while model robustness was confirmed through cross-validation. Given the difficulties of obtaining high-quality mental health data, this work relied on a publicly accessible dataset from www.kaggle.com, which contained known imbalances and incomplete entries. Empirical results showed that ensemble stacking classifiers reliably outperformed individual models, reaching an accuracy approaching 98%. Ultimately, this research seeks to refine mental health assessment practices and treatment approaches by harnessing advanced machine learning techniques to deepen understanding of these disorders.
