A Tailored Artificial Intelligence Model for Predicting Heat Strain of Construction Workers

Abstract

Abstract Traditional multilayer perceptron models have been used to predict the heat strain of construction workers. A major drawback of these models is that their black box nature may generate predictions that contradict domain knowledge and this casts doubt to construction managers on the effectiveness of the models. To overcome this limitation, a tailored multilayer perceptron model is developed to predict the heat strain of construction workers that guarantees the monotonicity of the predicted heat strain with some input features (e.g., temperature). The main finding is the tailored multilayer perceptron model never predicts results that contradict domain knowledge, making it more acceptable to construction managers. The tailored multilayer perceptron model is validated by a Hong Kong based smart solutions company.