# Cooperative Evaluation of the Cause of Urban Traffic Congestion via Connected Vehicles

## TL;DR

This paper presents a distributed data mining system utilizing connected vehicle communication to accurately identify causes of urban traffic congestion, outperforming traditional methods in detection accuracy and speed.

## Contribution

It introduces a novel cooperative framework combining Voting Procedures, Belief Functions, and Data Association Techniques for better congestion cause estimation using connected vehicle data.

## Key findings

- Enhanced congestion cause estimation accuracy by up to 71%
- Reduced detection time by approximately 10%
- Fewer false alarms compared to Back-Propagation algorithm

## Abstract

We developed a distributed data mining system to elaborate a decision concerning the cause of urban traffic congestion via emerging connected vehicle (CV) technology. We observe this complex phenomena through the interactions between vehicles exchanging messages via Vehicle to Vehicle (V2V) communication. Results are based on real-time simulation generated scenarios extended from the real-world traffic Travel and Activity PAtterns Simulation (TAPAS) Cologne scenario. We evaluate a Voting Procedure (VP) useful for obtaining deeper insights using cooperation between vehicles, Belief Functions (BF) aim at improving representation of information and a Data Association Technique (DAT) aiming at data mining and extracting the association rules from the messages exchanged. Methods are tested and compared using a microscopic urban mobility simulator, SUMO and a network simulator, ns-2, for the simulation of communication between CVs. Compared to the Back-Propagation algorithm (BP) extensively used in the past literature, our performance evaluation shows that the proposed methods enhance the estimation of the cause of congestion by 48\% for the proposed VP, 58\% for the BF, 71\% for the DAT and 70\% for \textbeta-DAT. The methods also enhance detection time from 7.09\% to 10.3\%, and \textbeta-DAT outperforms BP by approximately 1.25\% less false alarms triggered by the network, which can be significant in the context of real-time decision making. We show that a market penetration rate between 63\% and 75\% is enough to ensure satisfactory performance.