Forecasting new diseases in low-data settings using transfer learning

Predicting Kyasanur forest disease in resource-limited settings using event-based surveillance and transfer learning

In recent years, the reports of Kyasanur forest disease (KFD) breaking endemic barriers by spreading to new regions and crossing state boundaries is alarming. Effective disease surveillance and reporting systems are lacking for this emerging zoonosis, hence hindering control and prevention efforts. We compared time-series models using weather data with and without Event-Based Surveillance (EBS) information, i.e., news media reports and internet search trends, to predict monthly KFD cases in humans. We fitted Extreme Gradient Boosting (XGB) and Long Short Term Memory models at the national and regional levels. We utilized the rich epidemiological data from endemic regions by applying Transfer Learning (TL) techniques to predict KFD cases in new outbreak regions where disease surveillance information was scarce. Overall, the inclusion of EBS data, in addition to the weather data, substantially increased the prediction performance across all models. The XGB method produced the best predictions at the national and regional levels. The TL techniques outperformed baseline models in predicting KFD in new outbreak regions. Novel sources of data and advanced machine-learning approaches, e.g., EBS and TL, show great potential towards increasing disease prediction capabilities in data-scarce scenarios and/or resource-limited settings, for better-informed decisions in the face of emerging zoonotic threats.

Deep learning modelling of public's sentiments towards temporal evolution of COVID-19 transmission

Public sentiments towards global pandemics are important for public health assessment and disease control. This study develops a modularized deep learning framework to quantify public sentiments towards COVID-19, followed by leveraging the predicted sentiments to model and forecast the daily growth rate of confirmed COVID-19 cases globally, via a proposed G parameter. In the proposed framework, public sentiments are first modeled via a valence dimensional indicator, instead of discrete schemas, and are classified into 4 primary emotional categories: (a) neutral; (b) negative; (c) positive; (d) ambivalent, by using multiple word embedding models and classifiers for text sentiments analyses and classification. The trained model is subsequently applied to analyze large volumes (millions in quantity) of daily Tweets pertaining to COVID-19, ranging from 22 Jan 2020 to 10 May 2020. The results demonstrate that the global community gradually evokes both positive and negative sentiments towards COVID-19 over time compared to the dominant neural emotion at its inception. The predicted time-series sentiments are then leveraged to train a deep neural network (DNN) to model and forecast the G parameter by achieving the lowest possible mean absolute percentage error (MAPE) score of around 17.0% during the model's testing step with the optimal model configuration.