Predicting the outcomes and impacts of occupational accidents in building construction utilizing machine learning and risk optimization

The construction industry is one of the sectors with the highest number of workplace accidents worldwide. In this context, raising awareness of occupational health and safety among stakeholders is crucial to preventing accidents and making the working environment safer. In this thesis, based on data from past workplace accidents in building construction, the root causes of accidents were examined in detail, and the aim was to separately estimate both the consequences (first aid, injury, death) and the 13 different impacts (increased costs, labor shortages, delays, etc.) of workplace accidents in the construction industry. Within the scope of the thesis, a risk assessment model was also developed using genetic algorithm optimization. With the developed model, the necessary measures to be taken against risks were identified, and cost-benefit optimization was performed. Two different approaches were adopted as data collection methods in the study. Based on studies in the literature, a questionnaire was created specifically for this thesis to collect data on accidents that have occurred. The 203 data obtained were statistically analyzed, and models predicting the outcomes and impacts of accidents were established using machine learning algorithms via the Python program. For the risk assessment model, sample risk assessment forms were created for specific work items, considering literature and previously developed risk assessment forms. Subsequently, risk scores for these work items were calculated by experts based on multi-story building constructions. Then, risk-cost optimization was achieved. The success metrics of the machine learning prediction models created in the first phase of the study were comprehensively presented in the study and compared with studies published in the literature. The results of the study show that the consequences and impacts of accidents can be predicted to a certain extent using machine learning algorithms. Thanks to the developed optimization model, a decision support model has been developed that can offer optimal combinations of measures that simultaneously provide minimum total risk score and minimum cost. It was also determined that the model is lower compared to the traditional risk assessment method. In addition, the study offers recommendations for the widespread use of artificial intelligence-based applications specific to the sector.