Dynamic feature selection improves influenza forecasting accuracy and generalization across countries
Summary AI-generated
- TL;DR
- A new machine learning model improves influenza forecasting by dynamically selecting the most relevant data features over time, proving highly accurate across different countries.
- Problem
- Predicting influenza outbreaks is challenging because the factors driving transmission change over time and vary significantly between regions. Traditional forecasting models often struggle to adapt to these shifting patterns, limiting their accuracy and their ability to work across different countries.
- Method
- The researchers developed AdaFluDR, a model that combines advanced neural networks with a dynamic feature selection mechanism. This mechanism continuously scores and selects the most relevant data inputs based on their timing and frequency patterns. A neural network then processes these selected features to generate influenza predictions for up to four weeks in advance.
- Results
- AdaFluDR outperformed traditional forecasting methods and other machine learning approaches across one- to four-week prediction windows. Furthermore, the model demonstrated strong generalization capabilities, maintaining high accuracy when applied to data from the United States, Canada, and Portugal.
- Takeaways
- Dynamic feature selection is crucial for handling the time-varying nature of epidemic drivers. The study shows that a single machine learning framework can generalize across different countries without needing complete restructuring. This approach offers a more reliable tool for mid-term epidemic forecasting.
- For industry
- For healthcare providers and public health organizations, this research offers a more adaptable forecasting tool that automatically adjusts to changing data trends without manual intervention. By accurately predicting flu trends up to a month ahead, organizations can better allocate medical resources and plan public health campaigns.
- Why it matters
- This work enhances global epidemic preparedness by providing a reliable, cross-border framework for infectious disease forecasting. Its ability to generalize across different national contexts makes it a valuable asset for international health agencies coordinating responses to seasonal outbreaks.
Abstract
INTRODUCTION: The differentiation in epidemic patterns and multiple influencing factors pose significant challenges to influenza forecasting, highlighting the need for novel methods to improve predictive accuracy and cross-regional generalizability. OBJECTIVES: This study aims to develop an adaptive feature selection model named AdaFluDR to address the time-varying nature of influenza transmission drivers across different periods and regions. METHODS: AdaFluDR integrates the SpaceTime and Crossformer models and utilizes a correlation-driven mechanism. This mechanism constructs a comprehensive score by integrating the temporal, frequency, and time domain information of features, and dynamically adjusts feature processing pathways based on this score. Subsequently, a multilayer perceptron (MLP) will be employed to model the nonlinear mapping relationship between the integrated features and the target variable for prediction generation. RESULTS: The AdaFluDR model outperforms traditional methods and other machine learning approaches, demonstrating robust predictive performance across multiple forecasting horizons (1-4 weeks), and strong generalization ability across the United States, Canada, and Portugal. CONCLUSION: Our study provides a novel and practical framework for forecasting influenza activity with reliable accuracy and cross-national applicability, providing a valuable tool for improving global epidemic preparedness and response strategies.