This is the code and processed data related to the publication:
Dong, Y., Zhang, L., Farah, H., Zgonnikov, A., & van Arem, B. (2023). Data-driven Semi-supervised Machine Learning with Surrogate Safety Measures for Abnormal Driving Behavior Detection. arXiv preprint arXiv:2312.04610. https://arxiv.org/abs/2312.04610
The original data is from https://github.com/UCF-SST-Lab/UCF-SST-CitySim1-Dataset
The codes make use of open-sourced implementation of HELM and other semi-supervised learning algorithms.
After setting up the folder and fetching the data, one can simply run the code with the specific function (identified by their names) get the relevant results.
Details about the implementation are demonstrated in the paper.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Detecting abnormal driving behaviour is critical for road traffic safety and the evaluation of drivers' behaviour. With the advancement of machine learning (ML) algorithms and the accumulation of naturalistic driving data, many ML models have been adopted for abnormal driving behaviour detection (also referred to as anomalies). Most existing ML-based detectors rely on supervised methods, which require substantial labelled data. However, ground truth labels are not always available in the real world, and labelling large amounts of data is tedious. Thus, there is a need to explore unsupervised or semi-supervised methods to make the anomaly detection process more feasible and efficient. To fill this research gap, this study analyzes large-scale real-world data revealing several abnormal driving behaviours (e.g., sudden acceleration, rapid lane-changing) and develops a Hierarchical Extreme Learning Machines (HELM) based semi-supervised ML method using partly labelled data to accurately detect the identified abnormal driving behaviours. Moreover, previous ML-based approaches predominantly utilized basic vehicle motion features (e.g., velocity and acceleration) to label and detect abnormal driving behaviours, while this study seeks to introduce event-level safety indicators as input features for ML models to improve detection performance. Results from extensive experiments demonstrate the effectiveness of the proposed semi-supervised ML model with the introduced safety indicators serving as important features. The proposed semi-supervised ML method outperforms other baseline semi-supervised or unsupervised methods regarding various metrics, e.g., delivering the best accuracy (99.58%) and the best F-1 measure (0.9913). The ablation study further highlights the significance of safety indicators for advancing the detection performance.