Short-term local predictions of COVID-19 in the United Kingdom using dynamic supervised machine learning algorithms

This study aimed to develop supervised machine-learning algorithms using multiple digital metrics (symptom search trends, population mobility, and vaccination coverage) to predict local-level COVID-19 growth rates in the UK. Dynamic supervised machine-learning algorithms based on log-linear regression were used to predict 1-week, 2-week, and 3-week ahead growth rates at the lower tier local authority (LTLA) level. Model performance was assessed using mean squared error (MSE), comparing the optimal models to naïve and fixed-predictors models. Real-time model performance was assessed at eight checkpoints between March 1st and November 14th, 2021. An online application, COVIDPredLTLA, was developed to visualize real-time predictions. The optimal models showed improved accuracy (21-35% MSE reduction) compared to naïve models, even during the Delta variant surge. Dynamic models demonstrated advantages over fixed-predictors models after several updates. The study concludes that the dynamic modeling framework shows promise in predicting short-term COVID-19 case changes, and the COVIDPredLTLA application could assist in decision-making for control measures and healthcare capacity planning.

Xin Wang, Yijia Dong, William David Thompson, Harish Nair, You Li