Incorporating weighted block sampling techniques in epidemic curve classification

In infectious disease outbreaks, transmission rate changes over time and often reaches a peak in the middle of the epidemic period. Individual-level models (ILMs) are commonly used to understand how diseases spread, and are typically fitted using a Bayesian MCMC framework, which can be very slow and computationally demanding. To speed up analysis, an efficient alternative is model-classification techniques, such as random forests. However, standard random forests treat all predictor variables equally by randomly selecting split points, ignoring timing details, such as the peak of the outbreak. In this study, we propose using a block-sampling-based random forest that focuses on critical periods of the epidemic by assigning higher weights to data blocks that include the peak transmission time. This enables the model to focus more closely on the most critical aspects of the outbreak while maintaining a fast, data-driven approach. We validate this method using the U. K. 2001 foot-and-mouth outbreak and simulated data. The results show that this approach outperforms standard random forests in predicting epidemic generating models, supporting its applicability to real-time epidemic prediction and response planning.