Scholarlink

The quality of life is greatly impacted by mental health illnesses worldwide, making sophisticated diagnostic instruments that strike a compromise between precision and interpretability necessary. This paper uses the DASS-42 dataset (n = 39,775) to predict stress, anxiety, and depression using a unique machine learning architecture that uses a stacking ensemble of k-Nearest Neighbors, Decision Trees, and Multi-Layer Perceptron. In order to solve class imbalance and guarantee robust feature scaling, the model incorporates Yeo-Johnson transformations, Min-Max normalization, and NRS Boundary-SMOTE for data balancing. Grid Search and Nested Cross-Validation are used to maximize hyperparameter tuning, resulting in 99.94% (depression), 99.92% (anxiety), and 99.92% (stress) accuracy. The framework bridges the gap between predictive capacity and transparency, making it a dependable tool for clinical mental health evaluations due to its high accuracy and interpretability. This method offers a scalable, reliable solution for mental health diagnostics and advances AI-driven psychological treatments, opening the door for early diagnosis and individualized therapy.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Ensemble Machine Learning-Based Approach to Predict Human Mental States with Optimized Feature Selection and Data Balancing

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Md. Tofael Ahmed Bhuiyan,
  • Shahriar Manzoor,
  • Khandaker Mohammad Mohi Uddin

摘要

The quality of life is greatly impacted by mental health illnesses worldwide, making sophisticated diagnostic instruments that strike a compromise between precision and interpretability necessary. This paper uses the DASS-42 dataset (n = 39,775) to predict stress, anxiety, and depression using a unique machine learning architecture that uses a stacking ensemble of k-Nearest Neighbors, Decision Trees, and Multi-Layer Perceptron. In order to solve class imbalance and guarantee robust feature scaling, the model incorporates Yeo-Johnson transformations, Min-Max normalization, and NRS Boundary-SMOTE for data balancing. Grid Search and Nested Cross-Validation are used to maximize hyperparameter tuning, resulting in 99.94% (depression), 99.92% (anxiety), and 99.92% (stress) accuracy. The framework bridges the gap between predictive capacity and transparency, making it a dependable tool for clinical mental health evaluations due to its high accuracy and interpretability. This method offers a scalable, reliable solution for mental health diagnostics and advances AI-driven psychological treatments, opening the door for early diagnosis and individualized therapy.