An epidemic’s course isn’t defined solely by case counts over time. Human actions and movement levels can alter its progression, posing challenges for accurate forecasting. A model trained only on recent infection trends may lose effectiveness when travel restrictions, control measures, or population mobility suddenly shift. This patented method integrates that dynamic context into epidemic predictions. It first decomposes daily new-case data with empirical mode decomposition (EMD), breaking the original signal into components. After noise reduction and secondary synthesis, LSTM networks forecast these components. A key innovation involves transportation data: passenger flows from high-speed rail, flights, and road transport are used as indicators of local control measures. This information helps set the time frame for epidemic predictions. Essentially, the model’s view of recent data isn’t static but adapts based on external evidence of human movement, guiding how much past information it uses to predict future trends. This approach addresses limitations of traditional models that rely heavily on case data correlations, which can lead to errors when interventions change transmission patterns. The patent aims to improve epidemic forecasts under varying control measures. This insight underscores that context influences which historical data remain relevant. Instead of expecting an AI to interpret each period in the same way, this method allows external factors to inform how much of the past the model considers, enhancing its ability to predict future developments.
Forecasting an epidemic when human behavior changes
Patent number: CN 115662651 A
Inventor(s): Z. Han, Z. Yang, Z. Zeng, W. Guan, and Z. Lin
Citation: Z. Han, Z. Yang, Z. Zeng, W. Guan, and Z. Lin, “一种基于交通网的 EMD-LSTM 疫情预测方法 [EMD-LSTM epidemic prediction method based on transportation networks],” China Patent Application CN 115662651 A, Jan. 31, 2023.
