Gradient boosting: A computationally efficient alternative to Markov chain Monte Carlo sampling for fitting large...

Area of Science:

• Epidemiology and Biostatistics
• Computational Biology
• Veterinary Public Health

Background:

• Disease forecasting relies on analyzing large spatio-temporal datasets, often using Bayesian models.
• Markov chain Monte Carlo (MCMC) methods are standard for these models but computationally intensive for large datasets.
• Computational burden increases with the spatio-temporal scale of the model.

Purpose of the Study:

• To develop a computationally efficient algorithm for fitting Bayesian spatio-temporal mixed effects binomial regression models.
• To compare the performance of a novel gradient boosting approach against optimized MCMC methods.
• To apply and evaluate the method for forecasting vector-borne diseases in domestic dogs.

Main Methods:

• Proposed a gradient boosting algorithm for Bayesian spatio-temporal mixed effects binomial regression.
• Applied the method to forecast Lyme disease, anaplasmosis, ehrlichiosis, and heartworm disease.
• Compared the gradient boosting approach with a computationally optimized MCMC algorithm using extensive real-world data.

Main Results:

• The gradient boosting approach was several orders of magnitude faster than the optimized MCMC algorithm.
• Both methods achieved similar mean absolute prediction errors for disease forecasting.
• The study successfully generated monthly disease forecasts across the contiguous United States.

Conclusions:

• Gradient boosting offers a computationally efficient alternative for fitting complex spatio-temporal regression models.
• This method significantly reduces computational burden for large-scale disease surveillance and forecasting.
• The approach demonstrates practical utility in predicting canine vector-borne diseases with high accuracy and speed.