Twitter Analysis of Road Traffic Congestion Severity Estimation

Road traffic congestion is one of the problems in large cities. While Twitter has become a major means for communication and message dissemination among Internet users, one question that arises is whether Twitter messages alone can suggest road traffic congestion condition. This paper proposes a method to predict traffic congestion severity level based on the analysis of Twitter messages. Different types of tweets data are used to construct a C4.5 decision tree model for prediction, including tweets from selected road-traffic Twitter accounts, tweets that contain road-traffic-related keywords, and geo-tagged tweets whose volume suggests large crowds in certain areas. In the training that used tweets data of one of the top congested roads in Bangkok, density of tweets has high information gain in classifying congestion severity level. Via a web application, the model can provide a traveler with an estimate of the congestion severity level of the road at every 30 minutes. Such advance congestion information can help the traveler to make a better travel plan. In an evaluation using 10-fold cross validation, the model also shows good performance.