In a recent review of Performance Measurement System (PeMS) data quality for the Connected Corridors project along the I-210 corridor, it was discovered that almost 10% of HOV loops along 30-miles of freeway were actually located in the mainline. Since mainline lanes are impacted more severely by congestion during peak periods than HOV lanes, they exhibit lower speeds during these times of day. As a result, it is possible that this erroneous configuration contributes to wrongful HOV lane degradation in performance reports.
The Connected Corridors team at PATH is working on a project for Caltrans to explore the application of machine learning techniques to improve the accuracy of measurements and therefore directly improve the quality of management decisions. The main goal is to develop automated means to identify configuration errors of HOV lanes.
The project includes wide range of data science and machine learning tools and techniques. Data pipeline needs to collect data, clean, preprocess and add required features. Ground truth data from the I-210 pilot project will be used to develop classification models for identifying the misconfigured labels. In addition, unsupervised techniques will be applied on District 7 data to detect anomalies.
View our work here.