Deep learning for supervised classification of spatial epidemics.

Area of Science:

• Epidemiology and Public Health
• Computational Biology
• Machine Learning Applications

Background:

• Timely containment of emerging epidemics is critical for public health.
• Statistical disease transmission models are essential for developing containment strategies.
• Traditional inference methods (e.g., Bayesian Markov chain Monte Carlo) are computationally intensive.

Purpose of the Study:

• To explore supervised machine learning methods for rapid inference in epidemic modeling.
• To compare the performance of different machine learning models, including deep learning, against traditional methods.
• To assess the effectiveness of these methods on simulated epidemic data.

Main Methods:

• Supervised statistical and machine learning, focusing on deep learning and random forest classifiers.
• Application to simulated epidemic data from two distinct swine farm populations in Iowa.
• Comparative analysis of model performance based on inference speed and accuracy.

Main Results:

• Machine learning approaches offer significantly faster inference compared to traditional methods.
• Random forest models performed well on denser simulated epidemic populations.
• Deep learning models demonstrated superior performance on sparser simulated epidemic populations.

Conclusions:

• Supervised machine learning, particularly deep learning, provides a computationally efficient alternative for epidemic modeling inference.
• Model selection (e.g., random forest vs. deep learning) should consider population density characteristics.
• These methods can accelerate the development of effective epidemic containment strategies.