Train Station Congestion Prediction Based on LSTM with Interventional Few-Shot Learning

Abstract

Congestion prediction for public transportation such as conventional lines or Shinkansen has been increasing in recent years for reducing the density of large-scale places such as stations and airports. In this paper, We propose a congestion forecasting method that uses causal inference technology, which combine the deep learning model for time serise and few-shot learning. By using this method, we achieve better congestion prediction accuracy and model explainability than traditional LSTM models while using small-scale learning data for a Japanese train station. We look forward to using causal inference techniques to extend the explainability of our models in fulture, which can be used to analyze the factors that affect the congestion prediction results.