Twitter-informed Prediction for Urban Traffic Flow Using Machine Learning

Abstract

The current traffic system requires short-term traffic forecasting to manage and control the traffic flow. Irregular traffic events, such as road closures, accidents, and severe weather, reduce the accuracy of data-driven predictive models. Social media platforms, particularly Twitter can significantly help to realize a real-traffic flow system by representing traffic events. Combining traffic data with information about road disruptions posted on Twitter can improve urban traffic parameter prediction. This paper proposes an urban traffic flow prediction by combining massive traffic, calendar, and weather data with related tweet posts. As a case study, the model is implemented on an urban traffic dataset extracted from the California Performance Measurement System (PeMS) in the USA. To provide a reliable and accurate prediction, the proposed model is evaluated with several machine learning methods. The results from the empirical study show that when Twitter features are combined with traffic, weather, and calendar features, the prediction accuracy is enhanced. As a result, we obtain around 89 percent, 95 percent, 93 percent, 91 percent, 91 percent, and 95 percent R-squared from AdaBoost regression, Random Forest, Gradient Boosting, Artificial Neural Network, Decision Trees, and KNN Regression, respectively.