Steering a Historical Disease Forecasting Model Under a Pandemic: A Case of Flu and COVID-19
Forecasting influenza in a timely manner aids health organizations and policymakers with adequate preparation and decision making. Traditionally, flu forecasting efforts were aimed at predicting influenza-like-illness (ILI) case counts, which are defined based on symptomatic presentation. The COVID-19 pandemic has complicated forecasting efforts first because of similarities in symptoms and due to shifts in healthcare seeking behaviors of the general population. We term the ILI values observed when it is potentially affected by COVID-19 healthcare seeking patterns as COVID-ILI. Under the current pandemic, historical influenza models carry valuable expertise about the disease dynamics but face difficulties in adaptation. Therefore, we propose a neural transfer learning framework which allows us to 'steer' a historical disease forecasting model to new scenarios where flu and COVID co-exist. Our framework enables this adaptation by automatically learning when it should emphasize learning from COVID-related signals and when it should emphasize learning from the historical model. In this way, we exploit representations learned from historical ILI data as well as the limited COVID-related signals.
Challenges in accurately forecasting COVID-ILI include data paucity and complex dynamics. Our experiments demonstrate that our proposed framework is successful in adapting a historical forecasting model to ILI forecasting in the context of the current pandemic. In addition, we show that success in our primary goal, adaptation, does not sacrifice overall performance as compared with state-of-the-art influenza forecasting approaches. Joint work with: Alexander Rodriguez, Anika Tabassum, Dr. Bijaya Adhikari, Dr. Naren Ramakrishnan and Dr. B.Aditya Prakash.
Nikhil Muralidhar is a PhD student in the Computer Science department at Virginia Tech and is affiliated with the Discovery Analytics Center (DAC). He is advised by Dr. Naren Ramakrishnan. His work focuses on time series forecasting, anomaly detection and more recently on incorporating prior domain knowledge into machine learning models to enable learning under data paucity and in noisy data contexts. Nikhil is an NSF Urban Computing Fellow and serves as a PC member for ICML, ICLR, TKDE, IEEE Big Data. He has published in venues like IJCAI, SIAM SDM, IEEE Big Data and journals like ACM TIST and Big Data Journal.