Deep neural network analysis of coronavirus disease historical cases

COVID-19 represents a contemporary pandemic driven by highly contagious severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2. Nigeria had a share in this pandemic as the disease ravaged most parts of the country, leading to serious health and humanitarian crises. This paper presents the findings of an analysis of historical cases of coronavirus disease 2019 (COVID-19) from Nigeria, utilising a deep neural network (DNN). Four cases from the COVID-19 dataset were collected and formulated as time series, and a day-ahead prediction was conducted. The Levenberg-Marquardt algorithm was adopted for the update of the model’s parameters. The performance of DNN and Neural network (NN) with a single hidden layer was evaluated and compared. Mean Squared Error (MSE) and Mean Absolute Error (MAE) were the performance metrics employed. For daily cases, DNN exhibits reduced MSE and MAE values of 0.0358 and 0.01365 compared to NN with MSE and MAE values of 0.0580 and 0.1812, respectively. Similarly, for daily death, DNN has an MSE of 0.0274 and MAE of 0.1171, compared to NN, which has an MSE of 0.0364 and MAE of 0.134. DNN with MSE of 0.0155 and MAE of 0.0760 in total case performed better than NN with MSE of 0.0264 and MAE of 0.1116. Similarly, in total death, DNN has an MSE of 0.0001 and MAE of 0.0067 compared to NN with an MSE of 0.0015 and MAE of 0.0238. For the combined cases, DNN performed better than NN with MSE and MAE of 0.0017 and 0.0095 against 0.0078 and 0.0207 of NN, respectively. The results demonstrate that the deep neural network (DNN) outperforms the single-hidden-layer neural network (NN) in predicting COVID-19 cases and deaths in Nigeria, as evidenced by lower MSE and MAE values across all datasets. These findings highlight the potential of DNN for accurate time-series forecasting of infectious disease trends, which can aid in better decision-making and resource allocation during pandemics.