Modeling the behavior of construction workers to predict their propensity for unsafe acts

Abstract

The construction sector is leading in the number of accidents caused by the unsafe behavior of workers. Unsafe behavior can arise from a worker’s personal preferences or from any external unsafe condition in his working environment. Personal preferences are the intentional unsafe practices of workers and various behavioral constructs drive these preferences. On the other hand, the absence of no measures and controls for unsafe behaviors at the organizational level can be treated as an external unsafe condition that leads to unsafe acts. Therefore, this study proposes a propensity prediction engine powered by the classification algorithm of Artificial neural networks (ANN). The ANN-based propensity model takes quantified values of individual features of behavior-modifying constructs as inputs and provides outputs in the form of classification of workers as safe or unsafe. The behavioral constructs are taken from the theory of planned behavior (TPB) and the individual features of these constructs are explored from previous studies and field surveys. The model is a multi-layer feed- forward network with back-propagation built on an architecture of 10-16-6-2 and has been trained, validated, and tested using Keras API of Tensorflow. The study also presents a framework for practical implementations of propensity prediction engine for construction organizations. Specialized behavior interventions are proposed to be included in safety training programs for worker’s classified as unsafe by the prediction engine. The engine will help construction organizations in improving their safety training program by providing a way of managing behabehavioragement gaps at the organizational level.