Human-Centered Stress Monitoring in Smart Assembly Lines Using Explainable AI With SHAP

In manufacturing process automation and systems optimization, worker health and well-being face increasing challenges, with stress monitoring becoming crucial, particularly on assembly lines. This study presents a novel stress-monitoring framework designed for high-pressure manufacturing environments, seamlessly integrating IoT, AI, and advanced data analytics. Leveraging data from IoT-enabled wearable devices, we developed a robust machine learning model. A rigorous methodology guided the development and evaluation of four classifiers: Random Forest (RF), K-Nearest Neighbors (KNN), Decision Trees (DT), and XGBoost, with RF proving most effective for our dataset. We initially selected 35 candidate models, establishing a baseline with an RF model due to its strength in handling data complexity. To assess generalizability, another candidate model was randomly chosen to simulate unseen data. A key component was the use of SHapley Additive exPlanations (SHAP) for feature engineering, enhancing model transparency by offering insights into feature importance and decision-making. Additionally, SMOTE addressed class imbalance, while Optuna fine-tuned hyperparameters for improved generalization. The best model achieved 97.40% accuracy (Random Forest) in distinguishing between stressed and non-stressed states. This study applied the advanced method toward Industry 4.0 alongside advancements in smart manufacturing, emphasizing the importance of worker productivity and well-being. As a result, this application of machine learning empowers manufacturers to enhance operational efficiency, reduce costs, and maintain a high standard of product quality.