Forecasting weekly influenza-like illness counts in chongqing using a temporal attention-based multisource network (TAM-Net)

Routine ILI surveillance can be complemented by internet search and meteorological data, but integrating these data streams with different temporal patterns is an important consideration for multi-horizon forecasting. This study developed and evaluated a Temporal Attention-Based Multisource Network (TAM-Net) for 1- to 4-week-ahead forecasting of weekly ILI case counts in Chongqing using ILI surveillance, Baidu Search Index, and meteorological data. Weekly ILI surveillance, Baidu Search Index, and meteorological data from Chongqing (2015–2024) were integrated and chronologically divided into training, validation, and test sets (70%, 15%, and 15%, respectively). Using a 26-week input window, TAM-Net employed branch-specific LSTM and TCN encoders to extract temporal features from ILI and exogenous data, respectively. Multi-head self-attention and forecast-horizon-wise adaptive gating were incorporated to refine temporal representations and dynamically adjust branch contributions across the 1- to 4-week forecasting horizons. TAM-Net achieved an overall RMSE of 1,418.06 ± 104.93 and an R² of 0.754 ± 0.037, demonstrating favorable forecasting performance compared with the evaluated baseline models. Ablation experiments highlighted the contributions of heterogeneous data sources and architectural components. The forecast-horizon-wise adaptive gating mechanism dynamically adjusted the contributions of historical ILI and exogenous information across different forecasting horizons, while rolling-origin validation further supported the model’s predictive performance. TAM-Net combines branch-specific temporal encoding with forecast-horizon-wise adaptive fusion for short-term ILI forecasting. However, its evaluation was limited to a single region and retrospective data; external and prospective validation is needed to establish its generalizability.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-21
DOI
https://doi.org/10.1038/s41598-026-81486-x
Primary Topic
Data-Driven Disease Surveillance
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Forecasting weekly influenza-like illness counts in chongqing using a temporal attention-based multisource network (TAM-Net)

Jiang Long, Linghao Ni, Bin Peng, Jiawei Zhou et al.
Scientific Reports
Data-Driven Disease Surveillance
article

Forecasting weekly influenza-like illness counts in chongqing using a temporal attention-based multisource network (TAM-Net)

Jiang Long, Linghao Ni, Bin Peng, Jiawei Zhou, Yi Yuan, Ju Wang, Jie Chen
article en

Abstract

Routine ILI surveillance can be complemented by internet search and meteorological data, but integrating these data streams with different temporal patterns is an important consideration for multi-horizon forecasting. This study developed and evaluated a Temporal Attention-Based Multisource Network (TAM-Net) for 1- to 4-week-ahead forecasting of weekly ILI case counts in Chongqing using ILI surveillance, Baidu Search Index, and meteorological data. Weekly ILI surveillance, Baidu Search Index, and meteorological data from Chongqing (2015–2024) were integrated and chronologically divided into training, validation, and test sets (70%, 15%, and 15%, respectively). Using a 26-week input window, TAM-Net employed branch-specific LSTM and TCN encoders to extract temporal features from ILI and exogenous data, respectively. Multi-head self-attention and forecast-horizon-wise adaptive gating were incorporated to refine temporal representations and dynamically adjust branch contributions across the 1- to 4-week forecasting horizons. TAM-Net achieved an overall RMSE of 1,418.06 ± 104.93 and an R² of 0.754 ± 0.037, demonstrating favorable forecasting performance compared with the evaluated baseline models. Ablation experiments highlighted the contributions of heterogeneous data sources and architectural components. The forecast-horizon-wise adaptive gating mechanism dynamically adjusted the contributions of historical ILI and exogenous information across different forecasting horizons, while rolling-origin validation further supported the model’s predictive performance. TAM-Net combines branch-specific temporal encoding with forecast-horizon-wise adaptive fusion for short-term ILI forecasting. However, its evaluation was limited to a single region and retrospective data; external and prospective validation is needed to establish its generalizability.

Scientific Reports
Chongqing Medical University (CN)
Data-Driven Disease Surveillance
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon