Makine öğrenme algoritmalarını kullanarak tahminsel modellerden preskriptif kaza önleme modelinin türetilmesi

The main drive upon which this study relies is to introduce a prescriptive accident prevention model to avoid work related accidents by predicting outcomes of work-related accidents in pipeline construction with using machine learning algorithms. In depth study on construction accidents is crucial due to construction being one of the most hazardous industry and being temporary in nature. To come up with a prescriptive accident prevention model, all incident reports from a pipeline project were analysed and a data set was prepared for 1,184 cases with attributes. These attributes consist of twenty-four immediate causes, eighteen root causes and three consequences that are nearmiss, asset or property damage and injury. A machine learning tool, RapidMiner, is used to predict outcomes of the cases for different data subsets by using eleven different ML algorithms. One of the machine learning algorithms, Deep Learning, was selected due to performing better in predicting outcome of complex data sets and in majority of twelve data sets. Model performance was attempted to be optimized with parameter optimization. It was concluded that predictive models with optimized parameters can predict accident outcomes better and a prescriptive accident prevention model can be presented thanks to these predictions. With a prescriptive model, it may be possible to provide a foresight about the root cause, immediate cause, or date and time of potential accidents. The causes of accidents can be eliminated; hence accidents can be prevented with these predictions having statistical basis.