Our model for predicting the trajectory of COVID-19 is data-driven. This means that what happens in reality has priority over what we think should happen.

The model is an aggregate of several component models. The best-performing components are neural networks, but the aggregate also has autoregressive models, decision trees, epidemiological models, among others.

The aggregation method is designed to emphasize different components in areas where they are strongest. The emphasis changes for different pandemic conditions and in different locations.

The project started as a class competition in the CS156 course at Caltech in the Spring of 2020. It continued as a summer research project in Professor Abu-Mostafa's lab with 26 undergraduate and graduate students, including the winners of the class competition (Rahil Bathwal, Pavan Chitta, Kushal Tirumala, and Vignesh Varadarajan). Here is an article that tells the story of the project.

The project has been possible through generous support of:
- The Clinard Innovation Fund
- The Caltech SURF office
- Mr. Charles Trimble and Ms. Liying Huang
- MapBox Community