APPLICATION OF GRADIENT-BOOSTING AND DEEP LEARNING MODELS ON INFECTIOUS DISEASE OUTBREAKS

Abstract

This study demonstrates the need for events on Internet news to limit the spread of infectious disease in sub-
Sahara Africa. Evaluating the quality of surveillance system for preventing future trends with Internet health news,
provides awareness and information in real-time. The dataset is modelled for weekly short-term outbreaks with
154,057,341 reported cases extracted from regional sub-Sahara Africa countries from 2010-2020. AdaBoost,
Extreme Gradient Boost (XGBoost), Convolutional Neural Network (CNN) and Long Short-Term Memory
(LSTM) are employed for the evaluation of the internet health data. These models are adopted with varying
Learning Rates (LR) to determine a better model for real-time predictions and more affected region. Comparative
analysis for disease trends and predictions are performed using Mean Square Error and Mean Absolute Error. The
results show there is spike in East region between 386 weeks, Central sub-Sahara has minimum values of 1.90e-
06 and 7.49e-03 for Mean Square Error and Mean Absolute Error, respectively. The results show that XGBoost
predicted more with low training time than CNN and LSTM, though, not affected with varying LR values while
CNN predicted better with large LR of 0.1 with high training time. In conclusion, XGBoost works better at vary
learning rate compared to deep learning models.