Article Preview
TopIntroduction
Many doubts have emerged in recent months about the potential links between BCG vaccination, tuberculosis infection and the spread of COVID-19. Some work (Gupta, 2020; M. Gursel & I. Gursel, 2020; Redelman-Sidi, 2020; Schaaf et al., 2020; Hegarty et al., 2020) has been published in this context without being conclusive. So, to keep digging into the matter is much needed now in order to develop a prediction model to better control the pandemic while trying to know the extent of the three factors’ possible implications even though it would still be quite difficult to prove. At the beginning of epidemic, stochastic models were widely used, because a small group of carriers had infected people randomly. After that, researchers turned their attention to deterministic models, which make possible to predict the emergence of infection peaks and to define different control strategies. Work, published in this context, touched on studies on a single country (Xinguang & Bin, 2020; Toshikazu, 2020; Qun et al., 2020; Anzai et al., 2020; Jung et al., 2020), the impact of one or more parameters on the evolution of contagion (Anzai et al., 2020), the comparison between the evolution of the current epidemic with the one of previous versions of the corona virus (McAleer, 2020), the risk estimation of fatal cases (Jung et al., 2020) etc. We suggest moving towards advanced artificial intelligence techniques to try and develop a model to predict infection and recovery instances for any selected country given many inputs including BCG vaccination and TB infections rate. Such a model can serve as a reference and a tool to inform the public health professionals, clinicians and decision-makers, enabling them to take coordinative and collaborative efforts to control the pandemic. Models can, also, help to better understand the virus by comparing predictions in different situation.
Several studies invoke the prominence of deep neural networks (DNNs) which surpass the performance of the previous dominant paradigm in diverse machine learning applications (Bouhamed & Ruichek, 2018; Hinton et al., 2012; Mohamed et al., 2012; Ciresan et al., 2010; Yu et al., 2011). Deep Learning is a set of machine learning methods allowing to model data with a high level of abstraction. It is based on articulate architectures of various transformations in the nonlinear space (Bouhamed & Ruichek, 2018; Bengio, 2009). It is also considered as part of the Big Data domain. Current interest for Deep Learning is, not only for its conceptual advances, but also for its technological advances. As a matter of fact, all the available solutions (in terms of models learning) can exploit the immense reservoir of power computing, established through actual modern computers, as well as requesting the main processor (CPU) and the graphic dedicated processors (GPU) (Bouhamed & Ruichek, 2018). A Big Data model can adapt with enormous volume of data and with enormous sequential treatment of numbers exceeding most powerful server capacities (Bouhamed & Ruichek, 2018; Zikopoulos & Eaton, 2011). Since prediction, in our context, depends on observations obtained at previous timings, our scope was more about predicting time sequences. The prediction of recovery cases also depends on the predicted numbers of infected cases, the prediction of deaths also depends on the predicted numbers of infected and recovered cases so the model we are trying to develop must also consider this overlap or dependency of predictions.