Comparative analysis of travel time prediction algorithms for urban arterials using Wi-Fi Sensor Data

Travel time is one of the elementary traffic stream parameters in both users' and transport planners' perspective. Conventional travel time estimation methods have performed out of sorts for Indian urban traffic conditions characterized by heterogeneity in transport modes and lack of lane discipline. Robust to these limitations, Media Access Control (MAC) matching is perceived to be a reliable alternative for travel time estimation. To assist with real-time traffic control strategies, this study aims at developing a reliable structure for forecasting travel time on Indian urban arterials using data from Wi-Fi/ Bluetooth sensors. The data collected on an urban arterial in Chennai has been used as a case study to explain the value of such data and to explore its applicability in implementing various prediction models. To this end, this study examines and compares three different machine learning algorithms k-Nearest Neighbour (k-NN), Random Forest (RDF), Naive Bayes, and Kalman filtering technique for prediction. The performance of each model is evaluated to understand its suitability.