Scholarlink

<p>HIV (Human Immunodeficiency Virus) represents a dangerous infectious disease that significantly impacts the economy and society in many countries. Accurately forecasting the spread of HIV is essential to monitor and provide early warning in cities where the risk of transmission is high. Current epidemic forecasting models often learn from spatio-temporal graph networks based on transmission between adjacent regions. However, these models do not align effectively with HIV surveillance activities, which mainly focus on tracking recorded cases. To improve current approaches, we introduce CaseNet, a distinctive model for HIV forecasting applying attention-based graph networks to learn transmission correlations from a city-wide graph constructed by an infection relation between case surveillance records. CaseNet capitalizes on spatio-temporal relationships and epidemiological factors within records to reveal hidden transmission patterns in a city, controlling the HIV epidemic occurs at the district level, facilitating the monitoring and early detection of outbreaks before they escalate to a city-wide level. We extract subgraphs corresponding to districts from the city-wide graph and apply graph pooling on subgraphs to forecast each district. At this stage, the nodes within the subgraphs are embedded by combining their specific information and transmission correlations among cases throughout the city. By adopting this approach, CaseNet improves district-level forecasts by integrating specific district information and transmission trends in the city. The research experimented with the dataset of 40,611 anonymized case surveillance records from Ho Chi Minh City, Vietnam. The findings demonstrate that CaseNet outperforms benchmark models, particularly by improving the accuracy of the short-term HIV forecast.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Casenet: a novel HIV forecasting model using attention-based spatio-temporal graph network leveraging city-wide infection correlations

  • 期刊论文

  • 发表日期：

    2026年3月5日

查看全文
  • Dat Thanh Pham ,
  • Duong Van Nguyen,
  • Thanh Tan Tran,
  • Viet Anh Nguyen

摘要

HIV (Human Immunodeficiency Virus) represents a dangerous infectious disease that significantly impacts the economy and society in many countries. Accurately forecasting the spread of HIV is essential to monitor and provide early warning in cities where the risk of transmission is high. Current epidemic forecasting models often learn from spatio-temporal graph networks based on transmission between adjacent regions. However, these models do not align effectively with HIV surveillance activities, which mainly focus on tracking recorded cases. To improve current approaches, we introduce CaseNet, a distinctive model for HIV forecasting applying attention-based graph networks to learn transmission correlations from a city-wide graph constructed by an infection relation between case surveillance records. CaseNet capitalizes on spatio-temporal relationships and epidemiological factors within records to reveal hidden transmission patterns in a city, controlling the HIV epidemic occurs at the district level, facilitating the monitoring and early detection of outbreaks before they escalate to a city-wide level. We extract subgraphs corresponding to districts from the city-wide graph and apply graph pooling on subgraphs to forecast each district. At this stage, the nodes within the subgraphs are embedded by combining their specific information and transmission correlations among cases throughout the city. By adopting this approach, CaseNet improves district-level forecasts by integrating specific district information and transmission trends in the city. The research experimented with the dataset of 40,611 anonymized case surveillance records from Ho Chi Minh City, Vietnam. The findings demonstrate that CaseNet outperforms benchmark models, particularly by improving the accuracy of the short-term HIV forecast.