Abstract:Traffic flow forecasting is a crucial task in intelligent transport systems. Deep learning offers an effective solution, capturing complex patterns in time-series traffic flow data to enable the accurate prediction. However, deep learning models are prone to overfitting the intricate details of flow data, leading to poor generalisation. Recent studies suggest that decomposition-based deep ensemble learning methods may address this issue by breaking down a time series into multiple simpler signals, upon which deep learning models are built and ensembled to generate the final prediction. However, few studies have compared the performance of decomposition-based ensemble methods with non-decomposition-based ones which directly utilise raw time-series data. This work compares several decomposition-based and non-decomposition-based deep ensemble learning methods. Experimental results on three traffic datasets demonstrate the superiority of decomposition-based ensemble methods, while also revealing their sensitivity to aggregation strategies and forecasting horizons.

What problem does this paper attempt to address?

### What problems does this paper attempt to solve? This paper aims to solve the problems of complex pattern recognition and model generalization in traffic flow prediction. Specifically, the authors focus on the following points: 1. **Capturing complex patterns**: - Traffic flow data has complex time - series characteristics, and traditional statistical and shallow machine - learning methods are difficult to effectively capture these complex patterns. - Although deep learning can better adapt to these complex patterns, it is prone to over - fitting, resulting in poor generalization ability. 2. **Improving the generalization ability of deep - learning models**: - To overcome the problem that deep - learning models are prone to over - fitting in traffic flow prediction tasks, the authors explore the deep ensemble - learning method based on decomposition. - These methods decompose the time series into multiple simpler signals, and then use deep - learning models for modeling and integration to generate the final prediction results. 3. **Comparing the effects of different ensemble - learning methods**: - Currently, few studies compare the performance of the decomposition - based ensemble method with the non - decomposition method that directly uses the original time - series data. - This paper experimentally compares several decomposition - based and non - decomposition - based deep ensemble - learning methods and evaluates their performance in traffic flow prediction tasks. 4. **Exploring sensitivity issues**: - The research also explores the sensitivity of the decomposition - based method to aggregation strategies and prediction time ranges. - The results show that the decomposition - based method is very sensitive to these factors, and different settings may significantly affect its performance. ### Main contributions 1. **Empirical research**: - Through experiments, it is proved that the decomposition - based ensemble method can improve the performance of deep - learning models more than baseline methods (such as traditional bagging and multi - resolution ensemble) in traffic flow prediction tasks. 2. **Effectiveness of optimizing aggregation strategies**: - The influence of optimized aggregation strategies on the decomposition - based method is explored, and the results show that these methods are very sensitive to aggregation strategies. 3. **Influence of input and prediction time ranges**: - The influence of the amount of input data and the prediction time range on the decomposition - based method is studied, and it is found that these methods do not always benefit from a large amount of data. ### Summary Through empirical research, this paper proves the superiority of the decomposition - based deep ensemble - learning method in traffic flow prediction tasks and reveals the sensitivity of these methods to aggregation strategies and prediction time ranges. This provides an important reference for future research, especially in how to design more effective ensemble - learning strategies.

An Experimental Study on Decomposition-Based Deep Ensemble Learning for Traffic Flow Forecasting

A Visual Analytics Approach for Traffic Flow Prediction Ensembles.

A Deep Ensemble Approach for Long-Term Traffic Flow Prediction

Daily Long-Term Traffic Flow Forecasting Based on a Deep Neural Network

Fuzzy Hybrid Framework with Dynamic Weights for Short‐term Traffic Flow Prediction by Mining Spatio‐temporal Correlations

Daily Traffic Flow Forecasting Through a Contextual Convolutional Recurrent Neural Network Modeling Inter- and Intra-Day Traffic Patterns.

An adaptive composite time series forecasting model for short-term traffic flow

Traffic flow prediction for highway vehicle detectors through decomposition and machine learning

A Deep Learning Framework About Traffic Flow Forecasting for Urban Traffic Emission Monitoring System

An Enhanced Ensemble-Based Long Short-Term Memory Approach for Traffic Volume Prediction

Traffic flow prediction by an ensemble framework with data denoising and deep learning model

Acting as a Decision Maker: Traffic-Condition-Aware Ensemble Learning for Traffic Flow Prediction

Hybrid Extreme Learning for Reliable Short-Term Traffic Flow Forecasting

Optimized Structure of the Traffic Flow Forecasting Model With a Deep Learning Approach.

Contrastive Learning-Based Adaptive Graph Fusion Convolution Network With Residual-Enhanced Decomposition Strategy for Traffic Flow Forecasting

Traffic Flow Forecast Through Time Series Analysis Based on Deep Learning

Traffic Flow Prediction With Big Data: A Deep Learning Approach

A deep learning hierarchical approach to road traffic forecasting

Deep Architecture for Traffic Flow Prediction

A Study of Deep Learning for Network Traffic Data Forecasting