Interpreting car-following behavior is crucial for enhancing traffic behavior analysis and advancing Connected Automated Vehicle (CAV) technique. Traditional models often emphasize the replication of a host vehicle's behavior based on data from its nearest preceding vehicle, overlooking the possible influence of multiple preceding vehicles. This paper presents an Autoencoder based Deep Learning (ADL) model designed to interpret car-following behaviors within CAV systems. By incorporating kinematic data from multiple preceding vehicles, the model offers a more comprehensive understanding of the factors influencing car-following dynamics. The proposed ADL framework integrates a bidirectional GRU neural network as the encoder with an attention-based GRU neural network as the decoder, enabling a robust end-to-end interpretation. The NGSIM dataset is used for training and validating the ADL model, which offers detailed car-following data involving multiple preceding vehicles. Experimental results reveal that the inclusion of data from multiple preceding vehicles enhances the model's capability to interpret and predict heterogeneous driving behaviors accurately. Additionally, the ADL model outperforms traditional car-following models in predicting simulated speeds and positions. © 2025 Elsevier B.V., All rights reserved.