Machine-Learning Approach for Estimating Passenger Car Equivalency Factors using Crowdsourced Data

Passenger car equivalency (PCE) factors are used by the Highway Capacity Manual (HCM) to convert truck volumes to equivalent passenger car volumes. This is important to consider the effects of trucks for freeway performance metrics such as demand volumes or density-based level of service (LOS). PCE factors are typically calculated using multi-class volumes collected from traffic sensors. However, this requires the installation of sensors to obtain PCE values, which can be costly and time-consuming, with relatively low spatial coverage. Therefore, this study proposes a novel approach to estimate PCE volumes using crowdsourced data with wide coverage and some open data sets. A multi-class self-attention timeseries stacked autoencoder with XGBoost inputs (TS-SAE-XGB) model is proposed to estimate passenger car volumes, truck volumes, and single unit truck ratios. These three estimated parameters are then used as inputs to the proposed PCE interpolation algorithm which estimates PCE values using HCM 7^th edition methods. For comparison, three other machine learning models are evaluated when used with the proposed PCE interpolation algorithm to estimate PCE values. The models were compared for accuracy using 11 loop detectors in the Phoenix metropolitan area as ground truth volume values with a spatial leave-one-out cross validation. The proposed TS-SAE-XGB model yielded the best accuracy, estimating PCE values with a MAPE of 8.29%, and heavy vehicle factors with a MAPE of 3.98%. The proposed method provides transportation professionals with a practical method of estimating PCE values for freeways where sensors are unavailable.