- This event has passed.
Time Forecasting of COVID-19 Signals: Challenges and Model Development
July 31 @ 4:00 PM - 5:30 PM IST
Title: Real-Time Forecasting of COVID-19 Signals: Challenges and Model Development
Speaker: Dr. Aniruddha Adiga, Research Scientist, Biocomplexity Institute at the University of Virginia
Host Faculty: Prof. Chandra Sekhar Seelamantula, EE, IISc
Venue: Multimedia Classroom (MMCR), Department of Electrical Engineering, Indian Institute of Science
Date & Time: July 31, 2023; 4 PM onward (Coffee will be served during the talk)
COVID-19 is the largest pandemic the world has seen with approximately 700 million confirmed cases, 8 million confirmed deaths pandemic to date, and unprecedented social, economic, and political impact. During the pandemic, we also observed an extensive development of computational and mathematical models to aid policymakers and response efforts. An essential use of such models is in early warning systems and forecasting of COVID-19 signals. Real-time forecasting of COVID-19 signals is a challenging problem due to data quality issues, nonstationarity of time series, evolving targets, behavioral adaptations, etc. It has been observed that under such circumstances, ensemble models consisting of a diverse set of model classes are a better choice than individual models.
In this talk, I will discuss our efforts toward the development of an ensemble model consisting of statistical, deep learning, and compartmental models and our participation in national-level collaborative forecasting efforts. Through these efforts we have observed that all classes of models are important, however, different model classes performed differently during various phases of the pandemic. Armed with this understanding, I will present a modification to the ensembling method to employ this phase information and use different weighting schemes for different phases to produce improved forecasts. However, predicting the phases of the time series is another challenge, especially when behavioral and immunological adaptations govern the evolution of the time series. I will discuss a phase prediction algorithm that employs auxiliary datasets and transfer entropy techniques. We evaluate our model’s performance with other models in the collaborative effort.
Biography of the speaker: