Abstract
This research evaluates the effectiveness of deep learning approaches for detecting aggressive driving behaviours through time-series data collected from inertial sensors. The publicly available Driving Behaviour Dataset allowed us to develop and test three neural network models including Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), and Gated Recurrent Unit (GRU) using tri-axial accelerometer and gyroscope readings. The data preparation involved z-score normalization followed by sliding window segmentation and one-hot encoding of behaviour labels. The evaluation of model performance relied on standard classification metrics which included accuracy alongside precision, recall, F1-score and confusion matrices. The LSTM architecture achieved the highest classification accuracy at 96% which demonstrates its ability to detect driving behaviour patterns across time. The research demonstrates how deep learning models especially sequence-based LSTMs, show promise for real-time driver behaviour recognition, which could lead to applications in intelligent transportation systems and advanced driver-assistance technologies.