Predictive Model for Incident Severity at Railway Construction Site Using Rapid Miner

Abstract

Railway construction sites are prone to accidents; to be worse, it involves fatality. This is because many factors cannot be controlled due to the hectic working environment. Therefore, this study explores the incidents that happened using machine learning (ML) to predict the severity of incidents at railway construction sites. The study uses Support Vector Machines (SVM), Decision Trees (DT), Deep Learning (DL), and k-Nearest Neighbors (k-NN) implemented in RapidMiner software to analyse data from railway construction. ML is used because of its capability to learn about the relationship between each factor and parameter of the incident, thus producing relevant predictions of severity incidents. The goal is to identify high-severity incidents, develop a predictive model, and compare the ML methods' performance using metrics like accuracy, precision, recall, and F1-score. A 70:30 training-testing data split will be used, and the results aim to identify the best ML method for predicting incident severity at railway construction sites. SVM and DL are better at predicting the severity of accidents due to their high precision, with both having a 0.91 score for precision. At the same time, DT is favourable for minimising missed critical accidents due to its high recall of 0.89. k-NN shows the most unfavourable performance among these machine learning. This study will serve as a benchmark for future railway projects, informing mitigation actions and procedures and providing a deeper understanding of potential incidents.