A Travel Time Prediciton with Machine Learning Algorithms

It is Friday afternoon. You have a blind date at San Francisco at 8:00pm. Since you don’t want to give a bad first impression, you plan to leave early enough not to be late for the appointment. However, the traffic on 101North on every Friday is real heavy and unpredictable. Although it takes only 50 minutes to get there on other weekdays, it might take from 1.5 hours to 3 hours. So, when should you leave for SF to minimize the wasted time and meet the appointment schedule? This question above is commonly encountered by travelers, so they want an accurate travel time prediction. Adding to this case, accurate travel time prediction is also essential for traffic information and transportation systems such as vehicular navigation systems in cars or traffic boards on the roads to provide efficient route information. Furthermore, with this information, we can prevent unnecessary traffic jams with such a scheme as load balancing. In this project, we try to predict a travel time from one place to another in a freeway using machine learning algorithms. There are several research papers dealing with this topic. Especially, J. M. Kwon et al [3] and X. Zhang et al [2] form a good starting point. These previous works provide several travel time prediction algorithms using linear regression. For these algorithms, inputs for linear regression are historical data of traveling times and current states of highways. However, the authors for these papers did not consider other important factors such as what day of the week and which season of the year. By considering additional important factors, we expect to improve the accuracy of prediction. The purposes of this project are twofold; first, we evaluate existing algorithms presented in relevant research papers; second, we develop new algorithms to enhance the accuracy of prediction. For evaluation and development of travel time prediction algorithms, we need several types of data including an average speed of each freeway for different times, days, and months. These data are available to public at Freeway Performance Measurement System website [1]. This website provides historical and real-time freeway data from freeways in the State of California collected by sensors installed on major highways.