Smart City Traffic Patterns Prediction Using Machine Learning

Abstract

Traffic affects every citizen’s life in many ways by how long it takes for him or her to travel from home to office, the air condition he or she inhales, the strain generated by traffic jams, sleep, and workouts induced by time spent in traffic. Since motorists cannot see the entire traffic system, the urban traffic system must be anticipated in order to sensitize residents about their mobility choices and the subsequent impact on the environment, as well as to implement smart transport system. The paper used five machine learning models: Bagging (BAG), K-Nearest Neighbors (KNN), Multivariate Adaptive Regression Spline (MARS), Bayesian Generalized Linear Model (BGLM), and Generalized Linear Model (GLM) to predict traffic pattern in a smart city. The dataset consists of 48,120 rows and 4 columns from which the weekday, year, month, date, and time were extracted. Analysis results show that increase in the number of junctions of the city can alleviate problem being faced on the road by commuters. The Root Mean Square Error (RMSE) of BAG, KNN, MARS, BGLM, GLM are 13.09, 9.23, 23.34, 8.7, and 8.6 respectively. Experimental results demonstrated that GLM attained minimal prediction error compared to other machine learning models such as BAG, KNN, MARS, and BGLM used in this study.