Urban Traffic Flow Prediction: A MapReduce Based Parallel Multivariate Linear Regression Approach

Urban traffic flow has the property of complexity, uncertainty, and time-varying, which bring large difficulty to real-time and accurately forecast the traffic flow for traffic control and route guidance. In this paper, according to the characteristics of existing traffic flow prediction model for long processing time and memory constraints, a parallel multivariate linear regression model was designed based on MapReduce to real-time predict traffic flow. The model is composed of three MapReduce process to estimate the regression parameters. Furthermore, we design and implement a series of experiments to verify the effectiveness of the proposed parallel multivariate linear regression model through empirical research. Experimental results show that the multivariate linear regression prediction model based on MapReduce has better performance in both speedup and scaleup, and suit for analysis and prediction of large-scale multi dimensional and time-series traffic data.