Driving Style Based Trajectory Prediction of the Surrounding Vehicles using LSTM & ARIMA in Autonomous Driving

Abstract

This research explores the application of advanced time-series forecasting models to predict vehicle trajectories based on driving styles. The study utilizes vehicle trajectory pairs obtained from the I-80 and US-101 freeways, extracted from the NGSIM dataset. Principal Component Analysis (PCA) is employed to simplify characteristic indexes, leading to the identification of three distinct driving styles: aggressive, moderate, and traditional. To facilitate predictive analysis, three datasets are created, each representing a unique driving style cluster. The research employs Long Short-Term Memory (LSTM) and Auto Regressive Integrated Moving Average (ARIMA) models to forecast future trends within each driving style. Evaluation metrics, including Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and R² value, assess the accuracy and reliability of the forecasting models. The LSTM model, with its capacity to model complex temporal dependencies, delivers impressive results with low error metrics and high R2 value. This research demonstrates the efficacy of LSTM models in accurately predicting trajectories based on various driving styles. This highlights the effectiveness of employing LSTM algorithms within our study, showcasing their capability to capture complex temporal dependencies inherent in diverse driving behaviors. This underscores not only the strength of the LSTM model itself but also the successful application of our research methodology in leveraging this algorithm to achieve precise trajectory predictions.