Accident Analysis on Construction Sites Using Data Mining and Natural Language Processing

Abstract

The issue of worker protection is of paramount importance in many nations. The construction industry is singled out as the most dangerous to work in. In addition to the human toll, construction accidents may have a devastating economic impact. Analysis of accidents is crucial for preventing such mishaps in the future and developing sound risk management strategies. Summary reports of fatality and disaster investigations are available for previous incidents in the construction sector. The information on construction industry accidents examines here using text mining and natural language processing (NLP) methods. To categories the origins of the incidents, the support vector machine (SVM), linear regression (LR), K-nearest neighbour (KNN), decision tree (DT), Naive Bayes (NB), and an ensemble model. In addition, the weight of each classifier in the ensemble model optimizes using the sequential-quadratic programming (SQP) technique. The optimized ensemble model achieved a higher weighted F1 score on average than the other models tested in this research. The outcome demonstrates that the suggested method is more resilient in low-support scenarios. In addition, we present an unsupervised chunking method to extract accident-causing everyday items using grammatical rules found in the reports. Foreign objects often cause construction accidents, so locating and removing them is crucial. We address the suggested approaches' shortcomings and provide recommendations for moving further.