Investigating Abnormal Transaction Detection of Manual Toll Collection (MTC) of Expressway

The expressway in Thailand is an alternate way to avoid traffic congestion. More than 4 million transactions are daily and about 2,000 abnormal transactions are detected. As per standard procedure and process, every abnormal transaction must be invested by a human working for the audit department. Currently, the transaction must go through a manual investigation and validation process one by one, which not only impacts to long lead time for completing each transaction but also required huge manpower for inspection. This implies additional major expenses subject to those abnormal transactions. This article aims to identify the features that lead to the root cause of abnormal transactions by using raw transaction data from 4 routes with the highest traffic in EXAT. There are three statistic models which are Decision Tree, Random Forest, and Extra tree with one monthly data for analysis. The result shows that there are 5 significant features: detection of vehicle type, human error, plaza location, axle detection, and dual tire detection. However, each correlative features have different rankings. These significant features bring appropriate technology to alleviate the problem of abnormal transactions.