Current browse context:
stat.ML
Change to browse by:
References & Citations
Statistics > Machine Learning
Title: Applying Deep Learning to Detect Traffic Accidents in Real Time Using Spatiotemporal Sequential Data
(Submitted on 15 Dec 2019 (v1), last revised 22 Dec 2019 (this version, v2))
Abstract: Accident detection is a vital part of traffic safety. Many road users suffer from traffic accidents, as well as their consequences such as delay, congestion, air pollution, and so on. In this study, we utilize two advanced deep learning techniques, Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRUs), to detect traffic accidents in Chicago. These two techniques are selected because they are known to perform well with sequential data (i.e., time series). The full dataset consists of 241 accident and 6,038 non-accident cases selected from Chicago expressway, and it includes traffic spatiotemporal data, weather condition data, and congestion status data. Moreover, because the dataset is imbalanced (i.e., the dataset contains many more non-accident cases than accident cases), Synthetic Minority Over-sampling Technique (SMOTE) is employed. Overall, the two models perform significantly well, both with an Area Under Curve (AUC) of 0.85. Nonetheless, the GRU model is observed to perform slightly better than LSTM model with respect to detection rate. The performance of both models is similar in terms of false alarm rate.
Submission history
From: Amir Bahador Parsa [view email][v1] Sun, 15 Dec 2019 06:49:43 GMT (601kb)
[v2] Sun, 22 Dec 2019 20:45:56 GMT (601kb)
Link back to: arXiv, form interface, contact.