Multisource Data Fusion for Real-Time and Accurate Traffic Incident Detection via Predictive Analytics
-
2023-04-01
-
Details:
-
Creators:
-
Corporate Creators:
-
Contributors:
-
Corporate Contributors:
-
Subject/TRT Terms:
-
Publication/ Report Number:
-
Resource Type:
-
Geographical Coverage:
-
Edition:Final Report - April 2023 (May 2021 – April 2023)
-
Corporate Publisher:
-
Abstract:The objectives of this study are to (1) identify data sets available to MassDOT that can be used for realtime incident detection; (2) investigate how data from different sources can be integrated to improve incident detection; and (3) develop guidance for establishing trigger points to alert Highway Operations Center (HOC) operators about incidents on the road. Speed data available through the Regional Integrated Transportation Information (RITIS) platform are used for developing two alternative strategies: (a) an Artificial Intelligence (AI) model using supervised learning based on Long Short-Term Memory (LSTM) and Variational Autoencoders (VAE) layers for classifying records as normal events or incidents, and (b) an empirical rule-based method using historical speeds to establish threshold values, below which an alarm is issued requiring the HOC operator’s attention. Results on the AI model and a verified incident data set indicate a False Alarm Rate (FAR) of 0.0069% and a detection rate of 91.70%. For the empirical rule-based model, a 30-day off-line “field-test” was conducted for June 2021. Most of the events recorded by the MassDOT HOC were detected, and for most of these events the detection time was well before the “SENT-ON” time recorded in the HOC incident database.
-
Format:
-
Collection(s):
-
Main Document Checksum:
-
Download URL:
-
File Type: