[Paper Review] AdaEnsemble Learning Approach for Metro Passenger Flow Forecasting

Shaolong Sun, Dongchuan Yang+2 more|arXiv (Cornell University)|Feb 18, 2020

Traffic Prediction and Management Techniques38 references4 citations

TL;DR

This paper proposes AdaEnsemble, an adaptive ensemble learning framework that integrates variational mode decomposition (VMD), SARIMA, LSTM, and MLP models to forecast metro passenger flows. By decomposing flow data into periodic, deterministic, and volatile components—each modeled with specialized algorithms—and recombining forecasts via a meta-MLP, the approach achieves state-of-the-art accuracy and robustness on Shenzhen metro data across multiple evaluation metrics.

ABSTRACT

Accurate and timely metro passenger flow forecasting is critical for the successful deployment of intelligent transportation systems. However, it is quite challenging to propose an efficient and robust forecasting approach due to the inherent randomness and variations of metro passenger flow. In this study, we present a novel adaptive ensemble (AdaEnsemble) learning approach to accurately forecast the volume of metro passenger flows, and it combines the complementary advantages of variational mode decomposition (VMD), seasonal autoregressive integrated moving averaging (SARIMA), multilayer perceptron network (MLP) and long short-term memory (LSTM) network. The AdaEnsemble learning approach consists of three important stages. The first stage applies VMD to decompose the metro passenger flows data into periodic component, deterministic component and volatility component. Then we employ SARIMA model to forecast the periodic component, LSTM network to learn and forecast deterministic component and MLP network to forecast volatility component. In the last stage, the diverse forecasted components are reconstructed by another MLP network. The empirical results show that our proposed AdaEnsemble learning approach not only has the best forecasting performance compared with the state-of-the-art models but also appears to be the most promising and robust based on the historical passenger flow data in Shenzhen subway system and several standard evaluation measures.

Motivation & Objective

• Address the challenge of forecasting highly variable and non-stationary metro passenger flow data.
• Overcome limitations of single-model approaches by leveraging complementary strengths of diverse algorithms.
• Develop a robust, adaptive ensemble framework that improves forecasting accuracy and stability.
• Demonstrate the effectiveness of component-wise decomposition and specialized modeling on real-world metro data.

Proposed method

• Apply variational mode decomposition (VMD) to decompose raw passenger flow time series into three distinct components: periodic, deterministic, and volatile.
• Use SARIMA to model and forecast the periodic component, capturing seasonal and cyclical patterns.
• Employ a long short-term memory (LSTM) network to learn and predict the deterministic component, modeling long-term dependencies.
• Train a multilayer perceptron (MLP) to forecast the volatility component, representing irregular fluctuations.
• Reconstruct the final forecast by combining the predictions of the three component models using a meta-MLP network.
• Optimize the entire AdaEnsemble framework end-to-end using historical passenger flow data from the Shenzhen metro system.

Experimental results

Research questions

• RQ1 Can an ensemble of specialized models outperform individual forecasting models in metro passenger flow prediction?
• RQ2 How effective is VMD in decomposing complex passenger flow time series into interpretable, predictable components?
• RQ3 Does component-specific modeling (SARIMA for periodic, LSTM for deterministic, MLP for volatility) enhance forecasting accuracy?
• RQ4 To what extent does the meta-MLP recombination stage improve overall forecasting performance compared to simple averaging?
• RQ5 How robust is the AdaEnsemble framework across different time scales and data conditions in real-world metro systems?

Key findings

• The AdaEnsemble model achieves the best forecasting performance among all compared state-of-the-art models on Shenzhen metro data.
• The model demonstrates superior robustness across multiple evaluation metrics, including MAE, RMSE, and MAPE.
• Component-wise decomposition via VMD significantly improves the interpretability and accuracy of downstream forecasting.
• The integration of SARIMA, LSTM, and MLP for different components leads to more accurate predictions than using any single model alone.
• The meta-MLP recombination stage effectively combines forecasts, reducing error and enhancing generalization.
• Empirical results confirm that AdaEnsemble is not only accurate but also consistently reliable across diverse data conditions.