Missing data imputation for traffic flow based on combination of fuzzy neural network and rough set theory

Abstract

Currently, accurate traffic flow analysis and modeling are important key steps for intelligent transportation system (ITS). Missing traffic flow data are one of the most critical issues in the application of ITS. In this study, a hybrid method combining fuzzy rough set (FRS) and fuzzy neural network (FNN) is proposed for imputation of missing traffic data. Firstly, FNN is used for data classification, then the K-Nearest Neighbor (KNN) method is used to determine the optimal number of data used to estimate missing data in each category, and finally the fuzzy rough set is used to impute missing values. In order to validate the imputation performance of the proposed hybrid method, the traffic flow data collected from the loop detectors at different time intervals on roadway network are used in model calibration and validation. Three common indicators, including RMSE (root mean square error), R (correlation coefficient) and RA (relative accuracy), are used to evaluate the imputation performance under different data missing ratios. A model comparison is conducted between proposed imputation method and several widely used models including average-based and regression-based methods. The results show that the proposed method is superior to the traditional method for the traffic flow data collected at different time intervals with different missing ratios, which also further demonstrate its effectivity and validity.