Autoencoder and Multi-Head Attention with GRU based Approach to Predict Disease Outbreak using News-Crawl 2019 Data

Predicting disease outbreaks from news headlines is still challenging, as predictions rely on particular disease news data such as Ebola, Zika, SARS, MERS, COVID-19, and many more. A combined approach has been presented in this research work, where an Autoencoder and Multi-Head Attention combined with GRU used to predict disease outbreak. This work utilizes the News-Crawl 2019 data, where generic keywords are used to extract disease outbreak news headlines to predict whether or not news headlines are related to disease outbreaks. The work gains significance as the prediction is done without any bias, i.e., without being specific to the names of the disease to be predicted. The proposed approach achieves good accuracy on the test data through performance evaluation and has the capability to predict any disease outbreak, rather than just specific disease outbreaks, from news data. Whereas the combination of Autoencoder and Multi-Head Attention with GRU achieves 97.96% accuracy on test data.