Machine learning based algorithm for prediction of Covid-19 cases

Abstract

The global pandemic caused by the SARS-CoV-2 virus, which resulted in the emergence of Covid-19, has required a shift in research methodologies toward effective identification, analysis, and control. This research focuses on the daily forecasting of Covid-19 cases in Iraq, using a large dataset that includes daily recorded cases as well as socio-demographic and health-related attributes. In the face of the ongoing pandemic, policymakers and medical authorities require reliable forecasting techniques to make informed decisions and develop robust strategies. However, forecasting accuracy is hampered by the dynamic nature of virus spread, inherent uncertainties, and the need to analyze large datasets. Artificial neural networks, convolutional neural networks (CNNs), long short-term memory networks (LSTMs), and a proposed Enhanced Hybrid CNN-LSTM model (EH-CNN-LSTM) are used in the research. Before model implementation, the dataset is preprocessed to address seasonality, residuals, and trends. The EH-CNN-LSTM model is an approach to forecasting Covid-19 cases that takes advantage of the strengths of both CNNs and LSTMs. The abstract emphasizes the difficulties associated with using CNN and LSTM models, as well as how the hybrid model overcomes these difficulties. The combination of CNNs allows for effective feature extraction from spatial data, while LSTMs capture temporal dependencies, resulting in a more accurate forecasting tool. The research shows that increasing the volume of training data improves predictive performance. When trained on 80% of the data and evaluated on 20% of the data, the EH-CNN-LSTM model outperforms other models, achieving a Mean Absolute Percentage Error (MAPE) of 5.28, a Mean Squared Logarithmic Error (MSLE) of 0.00, and a Root Mean Squared Logarithmic Error (RMSLE) of 0.02. Finally, the abstract emphasizes the importance of the proposed model in providing accurate daily forecasts for Covid-19 infection cases in Iraq.