Model for Filtering the Outliers in DSRC Travel Time Data on Interrupted Traffic Flow Sections

This study aims to develop a model that effectively filters outliers in travel time data obtained by using the DSRC on interrupted traffic flow sections. To establish a direction for model development, we identified the characteristics of the DSRC system and existing abnormal data filtering models, causes of outliers in travel time data and their distribution center, changes in the travel time distribution according to the section length and existence of traffic congestion, variation in the number of data samples obtained, and the status of missing data. When an existing model is used to filter the abnormal travel time data obtained on the interrupted traffic flow sections, the mean and median values derived based on the asymmetry of the travel time distribution deviate from the center of the travel time distribution, which leads to fundamental errors in the estimation of the size of the normal travel time data distribution. Moreover, the performance of this model for filtering abnormal data decreases because of the distorted mean and standard deviation when outliers are mixed and optimizing filtering variables is difficult. To overcome these limitations of existing outlier filtering models, we developed a novel outlier filtering model by establishing a model development direction to improve the methods of obtaining data, setting the travel time distribution center, estimating distribution size, and filtering repetitive outliers. The performance of the developed model in filtering outliers is verified by comparing it with that of existing models for the intelligent transport systems installed on 13 types of single-sections and multi-section on the metropolitan national highway, Korea. The comparison indicates that the developed model generally exhibits the best performance for filtering abnormal data on all types of sections. The normal filtering ratio of the developed model was maintained at over 98.5% for all road types and traffic situations, demonstrating improvement of up to 27.4%. The error ratio of travel time at short interrupted traffic flow sections during non-congestion is significantly improved in this model.