GPS Data for Behaviour Style Prediction in Driving Using Optimized AI Model

• Published in: 2023 International Conference on Data Science and Network Security (ICDSNS) pp. 1 - 8
• Main Authors: B D, Parameshachari, Srinivas, Chetana, Shruthi H M, Leena, D S, Manasa, Kumar, Keerthi, P, Rashmi
• Format: Conference Proceeding
• Language: English
• Published: IEEE 28.07.2023
• Subjects: Adaptation models; Deep learning; Dimensionality reduction; Feature extraction; Global Positioning System; Intelligent Transportation Systems; Lion Optimization Algorithm; Long short-term memory; Predictive models; Roads; Sensors; Spatiotemporal attention
• DOI: 10.1109/ICDSNS58469.2023.10245411

When a driver's actions while behind the wheel are observed and analysed in real time, the process is called driver profiling. Better transport safety and the advancement of Intelligent Transportation Schemes can be attained through a deeper comprehension of driving behaviour (ITS). Due to the lack of consideration for environmental factors (such as weather and traffic conditions), risk predictions based only on the inclusion of identified behaviours may be inaccurate. Personalized safety-based route planning is only one ITS application that can benefit from these profiles. Improved prediction accuracy has resulted from recent research that has used deep learning. But the inability to describe these models is a major drawback. This problem is addressed by the study's proposed solution: an LSTM perfect with spatial-temporal attention in driver behaviour prediction called spatiotemporal attention long short-term memory (STA-LSTM). Lion Optimization Algorithm is used to get the best value for the suggested model's weight (LOA). We apply the suggested model's optimization on 4032 observations from GPS sensors in Shenzhen, China. Two investigations found that drivers had a problem with maintaining the vehicle's lateral site, were speeding, and had irregular or rapid acceleration. Experiments have demonstrated the efficiency and accuracy of the proposed model, with low re-construction errors from large GPS datasets.