Aggressive Driving Behavior Detection Using Integrated I-DBSCAN and LSTM Neural Network

Abstract Lately, a significant rise in traffic fatalities linked to aggressive driving behaviors has been noted, accentuating the imperative need for research in this domain. Hence, the detection of aggressive driving is increasingly advocated as a strategy not only to alert drivers about their perilous behaviors but also to potentially diminish the incidence of accidents. In parallel, the widespread adoption of driving simulators in the automotive sector has unveiled their profound advantages, especially in terms of repeatability in a controlled environment, helping researchers significantly reduce time and cost of development. Furthermore, driving simulators also provide a safe platform for testing new technologies and evaluating driver behavior in various scenarios. The article presents a method for identifying aggressive driving by analyzing vehicle dynamics data (such as speed, acceleration, and steering angle) collected from simulations in an urban setting using the software SCANeR™Studio. The algorithm employs Iterative Density-based spatial clustering of applications with noise, an unsupervised learning technique, to cluster aggressive driving maneuvers with sub-classification in terms of comfort and safety. Further, this labeled data is used to train a Bayesian optimization-based long short-term memory neural network, a pattern recognition model for the detection of driving behavior. The research findings confirm the model’s notable ability to recognize aggressive driving behaviors, as indicated by the confusion matrix and a F-score of 0.869, showing great potential to enhance road safety.