Accident prediction in construction using hybrid wavelet-machine learning

Abstract

Occupational accident rates in construction projects are usually higher than other industries in most countries, even though safety management systems are continuously improving. This study aims to contribute to the body of construction safety management by coupling discrete wavelet transform (DWT) and different machine learning (ML) methods to predict the number of occupational accidents using time series data. A dataset that consists of 393,160 occupational accidents recorded in Turkey between 2012 and 2020 was analyzed to predict the number of accidents for short-term, mid-term and long-term time periods, 1-day, 7-day and 30-day ahead, respectively. Model performances of stand-alone ML algorithms are improved with DWT, and hybrid waveletANN showed the best performance. A dynamic utilization plan was proposed to the field of safety management by introducing a new theoretical and practical framework. This study also aims to fill the gap in the literature related to time series prediction models in construction safety management.