Artificial Neural Network in Predicting Risk Exposure in Malaysian Shipyard Industry

Abstract

Risk exposure prediction is an important task in risk management and control. The efficiency of occupational safety and health (OSH) risk prevention depends on the accuracy of predicting risk exposure. In this study, a multilayer perceptron training using the backpropagation algorithm neural network was developed and presented for risk exposure prediction in the Malaysian shipyard industry. The data was collected from industrial shipyards in Malaysia via related government agencies to train the model and evaluate its performance. The data was pre-processed to ensure homogeneity. The artificial neural network (ANN) model used 10 influencing factors as inputs for risk exposure prediction: gender, age, occupation, workplace factors, activities involved, nationality, working hours, educational level, years of employment, and working zone. Several network architectures were developed, and the best model was selected for the risk exposure prediction of workers in the shipyard industry. Three evaluation metrics used for the selection of the best modal were mean square error (MSE), mean average percentage error (MAPE), and correlated of coefficient (R). The results showed that the ANN model, which has an accurate performance of 90.2250% with a coefficient of correlation of 91.375%, can accurately estimate the risk exposure of workers in the shipyard industry. Sensitivity analysis also revealed that input factors, such as working hours and workplace factors, have significant effects on OSH risk prediction. Therefore, they should be taken seriously when dealing with the risk exposure in the Malaysian shipyard industry.