Susuta Ryosuke, Yamada Kenta, Takayasu Hideki, Takayasu Misako
School of Computing, Institute of Science Tokyo, 4259 Nagatsuta-cho, Midori-ku, Yokohama 226-8502, Kanagawa, Japan.
Faculty of Global and Regional Studies, University of the Ryukyus, 1 Senbaru, Nishihara 903-0213, Okinawa, Japan.
Entropy (Basel). 2025 Jun 26;27(7):686. doi: 10.3390/e27070686.
This study investigates the feasibility of using GPS data and frequency of COVID-19-related blog words to forecast new infection trends through a linear regression analysis. By employing time series' trend decomposition and Spearman's rank correlation, we identify and select a set of significant variables from the GPS and blog data to construct two models: a fixed-period model and a sequential adaptive model that updates with each new wave of infections. Our findings reveal that the adaptive model more effectively captures long-term trends, achieving approximately 90% accuracy in forecasting infection rates seven days in advance. Despite challenges in forecasting exact values, this research demonstrates that combining GPS and blog data through a dynamic, wave-based learning model offers a promising direction for enhancing the forecasting accuracy of COVID-19 spread. This approach has significant implications for public health preparedness.
本研究通过线性回归分析，探讨利用GPS数据和与新冠疫情相关博客词汇的出现频率来预测新感染趋势的可行性。通过运用时间序列的趋势分解和斯皮尔曼等级相关性分析，我们从GPS和博客数据中识别并选择了一组重要变量，构建了两个模型：一个固定周期模型和一个随每一波新感染浪潮更新的顺序自适应模型。我们的研究结果表明，自适应模型能更有效地捕捉长期趋势，在提前七天预测感染率方面达到了约90%的准确率。尽管在预测精确值方面存在挑战，但这项研究表明，通过基于动态浪潮的学习模型将GPS和博客数据相结合，为提高新冠疫情传播预测的准确性提供了一个有前景的方向。这种方法对公共卫生准备具有重要意义。
