Neural Network - Based Predictive Analytics of Covid-19 in the Health Sector for Decision Support in Kenya

Abstract

COVID-19 spread rapidly across the globe and claimed many lives. As a result, it was declared a pandemic. Kenya, like most countries was negatively affected by the disease. In the face of COVID-19, predictive analytics plays a crucial role by providing the opportunity to proactively combat and curb the spread of the virus. This can be achieved by ensuring that hospitals are equipped with enough health resources such as ICU beds and ventilators. Statistical models and regression have been adopted in recent studies to try and predict the spread of the virus. Visualization dashboards have also been created, but they do not inform decision making. In this study, the development of a predictive system will aid in forecasting the spread of the virus. There is currently adequate data on COVID-19 cases since its onset. Visual representations of data on the other hand, are always easy to comprehend. This study is therefore focused on using neural networks as a machine learning algorithm to create a predictive model. Open-source data on COVID-19 confirmed cases in Kenya was collected, pre-processed, and used to train the model. MSE, MAE and RMSE were used as performance metrics to test the model which was then used to predict the expected number of COVID-19 cases within the next 60 days. These predictions were visualized onto Tableau dashboard against the COVID-19 infection cases and Health Resources available to contain the spread of the virus, hence providing an avenue for health resource planning and decision making. The study adopted LSTM neural network algorithm for univariate time series forecasting. Univariate time-series forecasting proved to be the best approach since the virus is quite unpredictable due to the low correlation between the multiple parameters and variables associated with it. The study also established that, merging the powerful aspects of visualization and predictive analytics enhances decision support. The system was also presented to the users upon completion and an acceptance response of 95% was received.