Comparative Analysis of Attention Mechanisms in Densely Connected Network for Network Traffic Prediction

Recently, STDenseNet (SpatioTemporal Densely connected convolutional Network) showed remarkable performance in predicting network traffic by leveraging the inductive bias of convolution layers. However, it is known that such convolution layers can only barely capture long-term spatial and temporal d...

Full description

Saved in:
Bibliographic Details
Main Authors: Myeongjun Oh, Sung Oh, Jongkyung Im, Myungho Kim, Joung-Sik Kim, Ji-Yeon Park, Na-Rae Yi, Sung-Ho Bae
Format: Article
Language:English
Published: MDPI AG 2025-06-01
Series:Signals
Subjects:
Online Access:https://www.mdpi.com/2624-6120/6/2/29
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849467269046861824
author Myeongjun Oh
Sung Oh
Jongkyung Im
Myungho Kim
Joung-Sik Kim
Ji-Yeon Park
Na-Rae Yi
Sung-Ho Bae
author_facet Myeongjun Oh
Sung Oh
Jongkyung Im
Myungho Kim
Joung-Sik Kim
Ji-Yeon Park
Na-Rae Yi
Sung-Ho Bae
author_sort Myeongjun Oh
collection DOAJ
description Recently, STDenseNet (SpatioTemporal Densely connected convolutional Network) showed remarkable performance in predicting network traffic by leveraging the inductive bias of convolution layers. However, it is known that such convolution layers can only barely capture long-term spatial and temporal dependencies. To solve this problem, we propose Attention-DenseNet (ADNet), which effectively incorporates an attention module into STDenseNet to learn representations for long-term spatio-temporal patterns. Specifically, we explored the optimal positions and the types of attention modules in combination with STDenseNet. Our key findings are as follows: i) attention modules are very effective when positioned between the last dense module and the final feature fusion module, meaning that the attention module plays a key role in aggregating low-level local features with long-term dependency. Hence, the final feature fusion module can easily exploit both global and local information; ii) the best attention module is different depending on the spatio-temporal characteristics of the dataset. To verify the effectiveness of the proposed ADNet, we performed experiments on the Telecom Italia dataset, a well-known benchmark dataset for network traffic prediction. The experimental results show that, compared to STDenseNet, our ADNet improved RMSE performance by 3.72%, 2.84%, and 5.87% in call service (Call), short message service (SMS), and Internet access (Internet) sub-datasets, respectively.
format Article
id doaj-art-7a164ecc9aa84f83aca39ecad1454bcf
institution Kabale University
issn 2624-6120
language English
publishDate 2025-06-01
publisher MDPI AG
record_format Article
series Signals
spelling doaj-art-7a164ecc9aa84f83aca39ecad1454bcf2025-08-20T03:27:43ZengMDPI AGSignals2624-61202025-06-01622910.3390/signals6020029Comparative Analysis of Attention Mechanisms in Densely Connected Network for Network Traffic PredictionMyeongjun Oh0Sung Oh1Jongkyung Im2Myungho Kim3Joung-Sik Kim4Ji-Yeon Park5Na-Rae Yi6Sung-Ho Bae7Department of Artificial Intelligence, Kyunghee University, 1732, Deogyeong-daero, Giheung-gu, Yongin-si 17104, Republic of KoreaDepartment of Artificial Intelligence, Kyunghee University, 1732, Deogyeong-daero, Giheung-gu, Yongin-si 17104, Republic of KoreaDepartment of Artificial Intelligence, Kyunghee University, 1732, Deogyeong-daero, Giheung-gu, Yongin-si 17104, Republic of KoreaHanwha, 188 Pangyoyeok-ro, Bundang-gu, Seongnam-si 13524, Republic of KoreaHanwha, 188 Pangyoyeok-ro, Bundang-gu, Seongnam-si 13524, Republic of KoreaHanwha, 188 Pangyoyeok-ro, Bundang-gu, Seongnam-si 13524, Republic of KoreaHanwha, 188 Pangyoyeok-ro, Bundang-gu, Seongnam-si 13524, Republic of KoreaDepartment of Artificial Intelligence, Kyunghee University, 1732, Deogyeong-daero, Giheung-gu, Yongin-si 17104, Republic of KoreaRecently, STDenseNet (SpatioTemporal Densely connected convolutional Network) showed remarkable performance in predicting network traffic by leveraging the inductive bias of convolution layers. However, it is known that such convolution layers can only barely capture long-term spatial and temporal dependencies. To solve this problem, we propose Attention-DenseNet (ADNet), which effectively incorporates an attention module into STDenseNet to learn representations for long-term spatio-temporal patterns. Specifically, we explored the optimal positions and the types of attention modules in combination with STDenseNet. Our key findings are as follows: i) attention modules are very effective when positioned between the last dense module and the final feature fusion module, meaning that the attention module plays a key role in aggregating low-level local features with long-term dependency. Hence, the final feature fusion module can easily exploit both global and local information; ii) the best attention module is different depending on the spatio-temporal characteristics of the dataset. To verify the effectiveness of the proposed ADNet, we performed experiments on the Telecom Italia dataset, a well-known benchmark dataset for network traffic prediction. The experimental results show that, compared to STDenseNet, our ADNet improved RMSE performance by 3.72%, 2.84%, and 5.87% in call service (Call), short message service (SMS), and Internet access (Internet) sub-datasets, respectively.https://www.mdpi.com/2624-6120/6/2/29spatiotemporal data predictionDenseNetattention mechanism
spellingShingle Myeongjun Oh
Sung Oh
Jongkyung Im
Myungho Kim
Joung-Sik Kim
Ji-Yeon Park
Na-Rae Yi
Sung-Ho Bae
Comparative Analysis of Attention Mechanisms in Densely Connected Network for Network Traffic Prediction
Signals
spatiotemporal data prediction
DenseNet
attention mechanism
title Comparative Analysis of Attention Mechanisms in Densely Connected Network for Network Traffic Prediction
title_full Comparative Analysis of Attention Mechanisms in Densely Connected Network for Network Traffic Prediction
title_fullStr Comparative Analysis of Attention Mechanisms in Densely Connected Network for Network Traffic Prediction
title_full_unstemmed Comparative Analysis of Attention Mechanisms in Densely Connected Network for Network Traffic Prediction
title_short Comparative Analysis of Attention Mechanisms in Densely Connected Network for Network Traffic Prediction
title_sort comparative analysis of attention mechanisms in densely connected network for network traffic prediction
topic spatiotemporal data prediction
DenseNet
attention mechanism
url https://www.mdpi.com/2624-6120/6/2/29
work_keys_str_mv AT myeongjunoh comparativeanalysisofattentionmechanismsindenselyconnectednetworkfornetworktrafficprediction
AT sungoh comparativeanalysisofattentionmechanismsindenselyconnectednetworkfornetworktrafficprediction
AT jongkyungim comparativeanalysisofattentionmechanismsindenselyconnectednetworkfornetworktrafficprediction
AT myunghokim comparativeanalysisofattentionmechanismsindenselyconnectednetworkfornetworktrafficprediction
AT joungsikkim comparativeanalysisofattentionmechanismsindenselyconnectednetworkfornetworktrafficprediction
AT jiyeonpark comparativeanalysisofattentionmechanismsindenselyconnectednetworkfornetworktrafficprediction
AT naraeyi comparativeanalysisofattentionmechanismsindenselyconnectednetworkfornetworktrafficprediction
AT sunghobae comparativeanalysisofattentionmechanismsindenselyconnectednetworkfornetworktrafficprediction