Enhancing Real-Time Safety Monitoring: The Critical Role of Hyperparameter Tuning in Machine Learning with Missing Data

Real-time monitoring applications often face challenges from erroneous or missing data (MD) due to hardware malfunctions, environmental interference, or communication errors, creating additional difficulty in accurate classification and prediction. This study evaluates the impact of hyperparameter tuning (HT) on the performance of machine learning models—Gaussian Naive Bayes (GNB), random forest (RF), multi-layer perceptron (MLP), support vector machine (SVM), decision tree (DT), XGBoost (XGB), logistic regression (LR), and linear support vector classification (LSVC)—under varying levels of MD. Results indicate that while HT improves performance for most models, the degree of improvement varies significantly. MLP and LR demonstrate robust accuracy gains, even with high MD levels, while DT and RF show sharp declines. SVM and LSVC provide moderate improvements but lack resilience at high MD levels. GNB shows negligible changes. These findings highlight the importance of model selection and preprocessing and the limitations of HT in addressing MD challenges.