decision tree algorithms

Despite other modes of transportation, trains move just along one dimension. However, trains inevitably change their track or move to the opposite track in railway stations and ports using switch systems. Switches are vital for better operation and seamless movement of trains. Furthermore, they are crucial for the safety of movement in tracks due to high derailment potentials at switches; therefore, all parts of switches need to be continuously monitored. An increasing number of accidents in railway systems is highly dependent on switch performance. According to the Islamic Republic of Iran Railways, 90 percent of railway accidents in Tehran stations occur on switches, from which 25 percent happen due to switch defects. Therefore, condition evaluation of switches is of significant importance. Research studies have not been sufficiently conducted on automated condition evaluation of switches. This paper aims to develop a robust automated approach to evaluate switch conditions to be able to measure switch defects. Having taken some pictures from various switches with fixed angles and distance from rails, an image processing technique is applied to determine defects. The first step of image processing is to preprocess the images to increase their quality. The second step is to indicate the type and severity of defects using different algorithms. A Graphical User Interface (GUI) is developed to develop a user-friendly tool to be able to load images, preprocess the images, measure defects, and report the health condition of switches. Finally, the outcomes are validated by applying ground truth, which ends up with high accuracy of approximation of 87 percent.

Nowadays, a significant part of goods and passengers are transported on suburban highways with mainly high-speed vehicles. Hence, these highways are very prone to accidents with different injuries. Due to the high fatality or severe physical/mental injury rates caused by car crashes, analyzing these accident-prone areas and identifying the factors affecting their occurrences is crucial. The specific objective of the study was to compare Chi-square Automatic Interaction Detector (CHAID), Classification and Regression Tree (CART), C4.5 and C5.0 decision tree data mining classification algorithms in building classification models for the fatality severity of 2355 fatal crash data records during 2007-2009 occurred in the roadways of 8 states in the USA. The results were evaluated using the accuracy metrics such as overall accuracy, kappa rate, precision, recall, and F-measure. The investigations confirmed that C5.0 had the best performance with the overall accuracy, and kappa rates of 94% and 92%, respectively. Additionally, classified fatality severity levels of the crashes were proposed for each algorithm to generate risk maps on the roads, to create potential accident risk spots. Decision tree models can be used for real-time data to find invariants in the tree over a period of time, which would be beneficial for policymakers.