Travel time prediction using random forest

• Main Author: Pranesh, Chaitra
• Other Authors: Justin Dauwels
• Format: Theses and Dissertations
• Language: English
• Published: 2019
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: http://hdl.handle.net/10356/78749
• Institution: Nanyang Technological University

Rapidly increasing vehicle congestion has been deteriorating the quality of life of people in urban areas of many developed and developing countries, including India. Caused mainly by rapid changes in urbanization, economy levels, vehicle ownership, and population growth, congestion leads to problems such as increased travel time, air pollution, and fuel use as well as decreased accessibility and mobility. In this regard, effective measures must be taken to avoid traffic jams, which will in turn lead to the sustainable development of the city. Travel time prediction plays an important role in reducing congestion. It is an important issue in the area of Intelligent Transport System (ITS) and Advanced Traveler Information System (ATIS). The transportation system becomes more efficient if there exists a system which accurately predicts travel time. The passengers can plan their trips and choose the best route, depending on the traffic conditions. Machine learning methods are gaining a lot of importance in travel time prediction. Since the traffic data is large, random forest algorithm can successfully handle this to provide accurate results. Random forest is a supervised and an ensemble learning method which can be used for both classification and regression. Multiple decision trees are built and merged together to get more stable and accurate prediction. The data collected by RTA, New South Wales, Australia for the Westbound line has been utilized. The performance of the random forest model is very high and the predicted travel time has high level of accuracy in terms of Mean Absolute Percentage Error (MAPE) compared to other traditional methods such as Support Vector Machine (SVM), historical average, and simple linear regression.