Predicting Choice of Filtering of Motorized Two Wheelers in Urban Mixed Traffic

Moving through lateral gaps defined by slower vehicles, or "filtering," is a common characteristic of motorized two wheelers (MTWs) in mixed traffic conditions. Despite its potential benefits such as reduced journey time and emissions, filtering is a critical maneuver, owing to the lesser conspicuity of MTWs and their almost non-existent physical protection from crashes. An in-depth knowledge of filtering behavior is necessary for various theoretical and practical applications. In an attempt to model this distinct driving characteristic of MTWs, this study investigates the filtering interactions of MTWs in urban mixed traffic conditions. Specifically, the behavioral differences between following and filtering interactions were investigated while considering the conditions found in heterogeneous traffic. The choice of filtering was modeled based on multiple factors including spatial parameters and classification parameters. This paper employs a utility-based binary logit model (BLM) and two machine learning (ML) based models: random forest (RF); and adaptive neuro fuzzy inference system (ANFIS) to predict the filtering choice of MTWs in urban mixed traffic. A comparative analysis revealed that, generally if not always, the ML based models (prediction accuracy of RF and ANFIS was 90.07% and 96.02%, respectively) performed better than utility-based models (prediction accuracy of BLM was 80.05%). Owing to the better performance measures of the ANFIS technique, it can be considered a powerful tool for predicting following and filtering choices of MTWs, useful for policy makers designing intelligent transportation systems and microsimulation models.