A Comparison of Multiple Machine Learning Algorithms to Predict Whole-Body Vibration Exposure of Dumper Operators in Iron Ore Mines in India

Abstract

Background: This study deals with some factors that influence the exposure of whole-body vibration (WBV) of dumper operators in surface mines. The study also highlights the approach to improve the multivariate linear analysis outcomes when collinearity exists between certain factor pairs. Material and Methods: A total number of 130 vibration readings was taken from two adjacent surface iron ore mines. The frequency-weighted RMS acceleration was used for the WBV exposure assessment of the dumper operators. The factors considered in this study are age, weight, seat backrest height, awkward posture, the machine age, load tonnage, dumper speed and haul road condition. Four machine learning models were explored through the empirical training-testing approach. Results: The bootstrap linear regression model was found to be the best model based on performance and predictability when compared to multiple linear regression, LASSO regression, and decision tree. Results revealed that multiple factors influence WBV exposure. The significant factors are: weight of operators (regression coefficient β=-0.005, p<0.001), awkward posture (β=0.033, p<0.001), load tonnage (β=-0.026, p<0.05), dumper speed (β=0.008, p<0.001) and poor haul road condition (β=0.015, p<0.001). Conclusion: The bootstrap linear regression model produced efficient results for the dataset which was characterized by collinearity. WBV exposure is multifactorial. Regular monitoring of WBV exposure and corrective actions through appropriate prevention programs including the ergonomic design of the seat would increase the health and safety of operators.