Performance and improvement of deep learning algorithms based on LSTM in traffic flow prediction

Abstract Existing traffic flow prediction research lacks adaptability to complex traffic scenarios and has limited prediction accuracy. This paper introduces an improved LSTM (Long Short-Term Memory) algorithm and sliding window technology to improve the accuracy and stability of traffic flow predic...

Full description

Saved in:
Bibliographic Details
Main Authors: Wei Xu, Eric Blancaflor, Mideth Abisado
Format: Article
Language:English
Published: Springer 2025-03-01
Series:Discover Applied Sciences
Subjects:
Online Access:https://doi.org/10.1007/s42452-025-06702-1
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849734700668551168
author Wei Xu
Eric Blancaflor
Mideth Abisado
author_facet Wei Xu
Eric Blancaflor
Mideth Abisado
author_sort Wei Xu
collection DOAJ
description Abstract Existing traffic flow prediction research lacks adaptability to complex traffic scenarios and has limited prediction accuracy. This paper introduces an improved LSTM (Long Short-Term Memory) algorithm and sliding window technology to improve the accuracy and stability of traffic flow prediction. This article discussed the enhancement of traffic flow (TF) prediction using a hybrid LSTM-BiGRU-Attention model aimed at improving the accuracy of traditional LSTM models. By integrating BiGRU (Bidirectional Gated Recurrent Unit) and Attention mechanisms, the proposed model captured complex TF patterns and long-term dependencies more effectively. The hybrid model is applied to Beijing urban road data with a time granularity (TG) of 10 min and a window size of 30 min, achieving an RMSE (root mean square error) of 4.478, an MAE (mean absolute error) of 3.609, and an R2 of 0.965. The LSTM-BiGRU-Attention model had similar prediction errors in TF during peak and off-peak periods and maintained high stability under different TF conditions. LSTM-BiGRU-Attention outperformed a single LSTM model and other benchmark models in multiple performance metrics. When the TG is 10 min, the window size of 30 min has the best performance, with 4.478, 3.609, and 0.965, respectively. The combination of LSTM, BiGRU, and Attention provides an effective solution for the development of intelligent transportation systems, which helps to achieve more accurate TF prediction and optimized management.
format Article
id doaj-art-8abc4a6ccb2d41729f3e770ff9fc8e43
institution DOAJ
issn 3004-9261
language English
publishDate 2025-03-01
publisher Springer
record_format Article
series Discover Applied Sciences
spelling doaj-art-8abc4a6ccb2d41729f3e770ff9fc8e432025-08-20T03:07:44ZengSpringerDiscover Applied Sciences3004-92612025-03-017411510.1007/s42452-025-06702-1Performance and improvement of deep learning algorithms based on LSTM in traffic flow predictionWei Xu0Eric Blancaflor1Mideth Abisado2College of Computing & Information Technologies, National UniversitySchool of Information Technology, Mapua UniversityCollege of Computing & Information Technologies, National UniversityAbstract Existing traffic flow prediction research lacks adaptability to complex traffic scenarios and has limited prediction accuracy. This paper introduces an improved LSTM (Long Short-Term Memory) algorithm and sliding window technology to improve the accuracy and stability of traffic flow prediction. This article discussed the enhancement of traffic flow (TF) prediction using a hybrid LSTM-BiGRU-Attention model aimed at improving the accuracy of traditional LSTM models. By integrating BiGRU (Bidirectional Gated Recurrent Unit) and Attention mechanisms, the proposed model captured complex TF patterns and long-term dependencies more effectively. The hybrid model is applied to Beijing urban road data with a time granularity (TG) of 10 min and a window size of 30 min, achieving an RMSE (root mean square error) of 4.478, an MAE (mean absolute error) of 3.609, and an R2 of 0.965. The LSTM-BiGRU-Attention model had similar prediction errors in TF during peak and off-peak periods and maintained high stability under different TF conditions. LSTM-BiGRU-Attention outperformed a single LSTM model and other benchmark models in multiple performance metrics. When the TG is 10 min, the window size of 30 min has the best performance, with 4.478, 3.609, and 0.965, respectively. The combination of LSTM, BiGRU, and Attention provides an effective solution for the development of intelligent transportation systems, which helps to achieve more accurate TF prediction and optimized management.https://doi.org/10.1007/s42452-025-06702-1Traffic flow predictionCity roadLong short-term memoryBidirectional gated recurrent unitHybrid modelLSTM-BiGRU-attention model
spellingShingle Wei Xu
Eric Blancaflor
Mideth Abisado
Performance and improvement of deep learning algorithms based on LSTM in traffic flow prediction
Discover Applied Sciences
Traffic flow prediction
City road
Long short-term memory
Bidirectional gated recurrent unit
Hybrid model
LSTM-BiGRU-attention model
title Performance and improvement of deep learning algorithms based on LSTM in traffic flow prediction
title_full Performance and improvement of deep learning algorithms based on LSTM in traffic flow prediction
title_fullStr Performance and improvement of deep learning algorithms based on LSTM in traffic flow prediction
title_full_unstemmed Performance and improvement of deep learning algorithms based on LSTM in traffic flow prediction
title_short Performance and improvement of deep learning algorithms based on LSTM in traffic flow prediction
title_sort performance and improvement of deep learning algorithms based on lstm in traffic flow prediction
topic Traffic flow prediction
City road
Long short-term memory
Bidirectional gated recurrent unit
Hybrid model
LSTM-BiGRU-attention model
url https://doi.org/10.1007/s42452-025-06702-1
work_keys_str_mv AT weixu performanceandimprovementofdeeplearningalgorithmsbasedonlstmintrafficflowprediction
AT ericblancaflor performanceandimprovementofdeeplearningalgorithmsbasedonlstmintrafficflowprediction
AT midethabisado performanceandimprovementofdeeplearningalgorithmsbasedonlstmintrafficflowprediction