Impact of varying datasets for prediction of COVID - 19 cases

• Main Authors: A Mohamed Zaki, Zakarya, Hassan Abdalla Hashim, Aisha
• Format: Article
• Language: English
• Published: International Journal of Science and Research (IJSR) 2024
• Subjects: TK7885 Computer engineering
• Online Access: http://irep.iium.edu.my/112335/1/112335_Impact%20of%20varying%20datasets.pdf; http://irep.iium.edu.my/112335/; https://www.ijsr.net/issue1.php?page=80&i=8&edition=Volume%2013%20Issue%204,%20April%202024

COVID - 19 has been identified as a global pandemic, and many experiments are applying various numerical models to anticipate the virus's likely growth under development. It is responsible for the emergence of the highly contagious illness. It is impacting millions of people throughout the globe. It has created a change in the research community's orientations for identification, analysis, and control via the application of different statistical and predictive modelling methodologies. These numerical models are examples of decision - making techniques that depend significantly on data mining and machine learning to create predictions based on historical data. In order to make smart judgments and create strong strategies, policymakers and medical authorities need reliable forecasting techniques. These studies are carried out on a variety of small scale datasets including a few hundreds to thousands of records. This study uses different sets of datasets consisting of COVID - 19 instances recorded on a daily basis in Iraq, together with socio - demographic and health related attributes for the region. The primary goal of this research is to see what is the impact of varying datasets for daily forecasting of COVID - 19 instances using deep learning forecasting tools. The predictive modeling for daily COVID - 19 infection cases involved using neural network architectures like enhanced hybrid model built using a Convolutional Neural Network and a Long Short - Term Memory network (EH - CNN - LSTM. Prior to the modeling, appropriate procedures were used to prepare the data and identify any seasonality, residuals, and trends. The model is trained and tested on various splits of the dataset. It is discovered that the higher the amount of training data, the better the predicted performance. Mean Absolute Percentage Error (MAPE), Mean Squared Logarithmic Error (MSLE), and Root Mean Squared Logarithmic Error (RMSLE) are used to evaluate the predictive performance.