Temporal Dependency Modeling in Lane- Changing Decisions Using Long Short-Term Memory Networks

Abstract Decision-making is a multifaceted event because the action sequences inherent in human decision-making are shaped by complex cognitive processes, including but not limited to beliefs, desires, intentions, and the ability to infer the mental states of others (theory of mind) Lin et al. (2022). These situations introduce substantial obstacles to accurately forecasting human decisions, especially when fundamental psychological mechanisms are not considered. Addressing this, this study explores the temporal evolution of lane-changing decisions among drivers through the application of sequence models. By examining how decision-making patterns develop when drivers encounter a series of scenarios or familiarise themselves with the task, this study provides insights into time-sensitive decision-making. This study presents an LSTM-based model for predicting lane-changing decisions in road scenarios. Utilising the strength of LSTMs in capturing temporal dependencies, our model demonstrates high accuracy in training and validation phases, suggesting robust generalisation and potential for real-world Advanced Driver Assistance Systems (ADAS)