Back

An Imputation-Based Approach for Augmenting Sparse Epidemiological Signals

Benefield, A. E.; Williams, D.; Nagraj, V.

2024-08-03 public and global health
10.1101/2024.07.31.24311314 medRxiv
Show abstract

Near-term disease forecasting and scenario projection efforts rely on the availability of data to train and evaluate model performance. In most cases, more extensive epidemiological time series data can lead to better modeling results and improved public health insights. Here we describe a procedure to augment an epidemiological time series. We used reported flu hospitalization data from FluSurv-NET, the National Healthcare Safety Network, and flu outpatient visits from ILINet to estimate a complete time series of flu hospitalization counts dating back to 2009. The augmentation process includes concatenation, interpolation, extrapolation, and imputation steps, each designed to address specific data gaps. We demonstrate the forecasting performance gain when the extended time series is used to train flu hospitalization models at the state and national level.

Matching journals

The top 5 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.