Document Type
Thesis
Date of Award
12-31-2019
Degree Name
Master of Science in Biology - (M.S.)
Department
Federated Department of Biological Sciences
First Advisor
Eric Scott Fortune
Second Advisor
Horacio G. Rotstein
Third Advisor
Kristen E. Severi
Abstract
Predicting seasonal variation in influenza epidemics is an ongoing challenge. To better predict seasonal influenza and provide early warning of pandemics, a novel approach to Influenza-Like-Illness (ILI) prediction was developed. This approach combined a deep neural network with ILI, climate, and population data. A predictive model was created using a deep neural network based on TensorFlow 2.0 Beta. The model used Long-Short Term Memory (LSTM) nodes. Data was collected from the Center for Disease Control, the National Center for Environmental Information (NCEI) and the United States Census Bureau. These parameters were temperature, precipitation, wind speed, population size, vaccination rate and vaccination efficacy. Temperature was confirmed as the greatest predictor for ILI rates, with precipitation providing a small increase in predictive power. After training, the model was able to predict ILI rates 10 weeks out. As a result of this thesis, a framework was developed that may be applied to weekly ILI tracking as well as early prediction of outlier pandemic years.
Recommended Citation
Amendolara, Alfred, "Predictive modeling of influenza in New England using a recurrent deep neural network" (2019). Theses. 1739.
https://digitalcommons.njit.edu/theses/1739
