Additive LOG transformation distributed feature embedding convolutional neural learning classifier for early COVID-19 prediction

The COVID-19 pandemic is a severe disease that has claimed many lives. It is crucial to reduce the mortality rate and take essential steps to provide suitable treatment. This allows the government to strategize and control the spread of the disease or at the very least, uplift the morale of the general public. To classify patients' input and their medical files, various learning methods have been introduced to facilitate COVID-19 prediction. However, due to the extensive dataset, it took a considerable amount of time to train the program, resulting in ineffective predictions, higher infection rates, increased spread, and elevated death rates. The main objective of this research is to accurately predict COVID-19 at an earlier stage and in less time using the additive log transformed distributive feature embedding time-dependent regressive convolutional neural learning classifier (ALTDFETRCNLC). Initially, patient files are collected as input for the dataset. The additive log ratio is transformed using one hot encoding to preprocess and normalize the input data. The Tversky similarity indexed stochastic distributive feature embedding technique is employed to select relevant features efficiently. Finally, the Levenberg-Marquardt convolutional neural learning classifier is utilized to classify COVID-19 predictions. This approach has significantly improved prediction accuracy and considers space complexity. Experimental evaluation is conducted using the proposed ALTDFETRCNLC technique and existing methods, utilizing the COVID-19 dataset with different metrics. The results demonstrate that the ALTDFETRCNLC technique outperforms contemporary and conventional works in terms of prediction accuracy, precision, recall, and F-measure, showing improvements of 4%, 4%, 3%, and 3% respectively. Additionally, the ALTDFETRCNLC technique achieves faster prediction times with an 8% improvement and reduces the error rate and space complexity by up to 8% and 9% compared to existing methods.