Scholarlink

This study investigates non-intrusive approaches to workplace stress detection using machine learning/deep learning techniques. Current methods rely heavily on physiological measurements (ECG, EDA, PPG) or behavioral monitoring which face implementation challenges in corporate environments. We propose an alternative approach utilizing psychosocial stressors and demographic variables to develop an explainable AI model for occupational stress detection. This approach addresses the practical limitations of current methods, which fail to discover underlying contributors to employee stress in the workplace. Two datasets were evaluated: The corporate stress dataset (CSDIW) with 50,000 records across 30 features and the HR Analytics Job Prediction dataset (HAJP) with 15,000 records across 10 features. Machine learning models including Logistic Regression, Naïve Bayes, Neural Networks, Autoencoders + XGBoost and ensemble methods (SVM, RF and XGBoost) were implemented in different feature spaces. Results indicate improved performance on HAJP dataset with and without feature engineering whereas models consistently underperformed on the CSDIW dataset. The soft voting-based ensemble classifier turned out the best performer achieving an accuracy of 0.86 and an f1 score of 0.84. Our findings suggest that while psychosocial features hold promise for non-intrusive stress detection in workplace, data quality and appropriate labelling remain critical challenges.

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

Workplace Stress Prediction Using Explainable AI: A Non-intrusive Approach with Psychosocial and Demographic Features

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Juswin Sajan John,
  • Shiju George,
  • C. Sharon Roji Priya

摘要

This study investigates non-intrusive approaches to workplace stress detection using machine learning/deep learning techniques. Current methods rely heavily on physiological measurements (ECG, EDA, PPG) or behavioral monitoring which face implementation challenges in corporate environments. We propose an alternative approach utilizing psychosocial stressors and demographic variables to develop an explainable AI model for occupational stress detection. This approach addresses the practical limitations of current methods, which fail to discover underlying contributors to employee stress in the workplace. Two datasets were evaluated: The corporate stress dataset (CSDIW) with 50,000 records across 30 features and the HR Analytics Job Prediction dataset (HAJP) with 15,000 records across 10 features. Machine learning models including Logistic Regression, Naïve Bayes, Neural Networks, Autoencoders + XGBoost and ensemble methods (SVM, RF and XGBoost) were implemented in different feature spaces. Results indicate improved performance on HAJP dataset with and without feature engineering whereas models consistently underperformed on the CSDIW dataset. The soft voting-based ensemble classifier turned out the best performer achieving an accuracy of 0.86 and an f1 score of 0.84. Our findings suggest that while psychosocial features hold promise for non-intrusive stress detection in workplace, data quality and appropriate labelling remain critical challenges.