Travel Time Prediction using Data Analysis and Performance Evaluation of different System Architectures

Abstract

This projects aims to use predictive analysis to counter one of the largest problems Hong Kong faces, growing traffic saturation level. While doing this, we look to investigate and evaluate different system architectures in order to implement a highly scalable, efficient and fault tolerant system. The implemented solution uses previously collected traffic speed map data made available by the Traffic Department to generate prediction models, which can used to predict the traffic saturation level at a given time and date. The accuracy of this model needs to be high, hence different models need to be tested and evaluated in order to find the best. The implemented solution boasts an prediction accuracy of 74 - 79% for all the months and supports batch/single prediction at the same speed and computational overhead. More importantly, in order to deal with a large amount of users, it's important the the system architecture is infinitely scalable. The flexible architecture can allow the project to grow beyond one country's boundaries and expand to multiple in the future. Further, the average response time of the system needs to be fast in order to deal with users proficiently and efficiently. The implemented solution does exactly that and is capable of handling approximately 1550 concurrent users per second with a low error rate (primarily network/pipe errors) and having and individual request response time of 200ms for batch predictions and 72ms for the single node traffic prediction which considerably outperforms production machine learning system architectures like PredictionIO on a single node. Finally, the implemented mobile application boast a clean and clear user interface completely inline with Google's Material Design specifications and allows users to enter their source, destination and travel date and time and plots the predicted traffic saturation level for each of the nodes and the user's route on google maps. Further, the application shows the user a list of the nodes whose traffic has been predicted and the travel time on that node.