MACHINE LEARNING MODEL FOR PREDICTIVE ANALYTICS OF RAIL FRACTURE OCCURRENCE IN INDONESIA

• Main Author: Nurul Hakim, Nafisah
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/50423
• Institution: Institut Teknologi Bandung

The train is a mode of mass transportation that is still quite in demand by Indonesians society today. This causes the urgency of ensuring the safety of good train use from the service provider. Even so, KNKT found that most of the train accidents were caused by infrastructure factors, one of which was a broken rail. A further examination of the related incidents gives the result that the incident rail fracture was caused by several things, including from management and organizational aspects. This research was conducted to provide a new approach on problem-solving methods in management and organizational aspects by applying machine learning to assess rail conditions. The assessment method referred to in this study is the use of a machine learning model to predict the probability of a broken rail at a point with the rail’s condition specification entered as input to the model. Model development is carried out using the CRISP-DM methodology and several modeling techniques. The models resulted from the use of different techniques will be evaluated using accuracy value and compared with each other to produce the most appropriate model. Model development is carried out using a dataset of rail fracture events from 2017 to 2019 accompanied by technical and operational details. The scope of the observation location in the study was limited to the railway operation areas in South Sumatra, especially DIVRE III and DIVRE IV. Evaluation results of each model carried out at the end of the study concluded that random forest is the most appropriate technique to be used in making analysis models for predicting broken rail probability based on the data used in this research.