TRCSE: a reliable and robust stacked ensemble model for passenger flow data reliability in autonomous public transport

View at publisher Summarise with VIRAT
Authors: Renhao Cao, Bin Yu, Yu Zhou
Cited by: 0
References: 60

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Abstract Autonomous Public Transport (APT) is currently being piloted in urban transportation systems, and bus automation serves as a critical scenario. A data trust crisis has emerged in real-time passenger flow prediction for automatic dispatching systems, posing a severe safety challenge. Existing methods still struggle to effectively guarantee data reliability. To address this problem, we propose TRCSE, a three-layer stacked ensemble model based on LSTM, XGBoost, and LightGBM. The model is trained on a fused dataset that combines Beijing bus passenger flow data and CICIDS 2017. TRCSE integrates a Conditional Variational Autoencoder (CVAE) with a robust loss function to guide base models toward robust regions. It also incorporates Mixup data augmentation and Transfer Learning to enhance generalization performance. Additionally, the model leverages Shapley Additive exPlanations (SHAP) to interpret internal decision-making processes and elucidate the origins of robustness. Evaluation results show that, compared with state-of-the-art baseline models, the original stacked model achieves an RMSE of 250.0370 with an average reduction of 34.63%. Its RMSE standard deviation reaches 121.8900, corresponding to an average reduction of 42.25%. The original model also obtains an Attack RMSE of 746.9072 with an average reduction of 24.57%. Notably, TRCSE further optimizes the anti-attack performance. It reduces the Attack RMSE to 701.7380, representing a 6.05% reduction compared with the original stacked model. SHAP analysis reveals that the contribution of XGBoost increases to 28.57%, making it the key driver of model robustness. This research validates the feasibility of TRCSE as a reliable and interpretable data benchmark for APT dispatching systems. It ensures prediction accuracy and stability in complex threat environments, and offers an innovative solution to the data trust crisis in future intelligent transportation.

Fields of Study & Subjects

Adversarial Robustness in Machine Learning Traffic Prediction and Management Techniques Transportation Planning and Optimization

Publication details

Journal / source
Engineering Research Express
Volume
8
Issue
7
Pages
075235–075235
Publication date
25 March 2026
Language
EN