A hybrid ANN-CNN model for predicting non-linear relationship of covid-19 cases based on weather factors

Abstract

With the global increase in the emergence of viral diseases, the most recent being the Coronavirus Disease 2019 (COVID- 19) in 2020-2021, it has decimated the world with little understanding of its history and the factors that influence its transmission dynamics. Weather significantly influences the spread of respiratory infectious diseases like influenza, yet the impact of weather on COVID-19 transmission in Nigeria remains unexamined and necessitates further clarification. This study presents and compares the results of six machine learning models, the developed Hybrid ANN-CNN, ANN, CNN, LSTM, LASSO, and Multiple Linear Regression models, aiming to predict the impact of weather factors on COVID-19 cases. The dataset used in this study includes daily datasets of Nigerian COVID-19 cases and seven weather variables collected from May 1, 2020, to April 30, 2021. The results indicate that the developed Hybrid ANN-CNN outperforms the remaining five models based on Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) for all cases. Specifically, for confirmed cases, the Hybrid ANN-CNN had an MAE of 0.0274, for recovery cases 0.0257, and for death cases 0.0425. Similarly, for RMSE, the developed Hybrid ANN-CNN had values of 0.0469 for confirmed cases, 0.0813 for recovery cases, and 0.0840 for deaths. This was followed by LASSO with an MAE of 0.01384 and CNN and LSTM with 0.1384 and 0.1385, respectively.