Biomedical COVID19 Data Analysis and Impacts for Post Event Using Machine Learning

Abstract

The COVID-19 public health pandemic, triggered by SARS-CoV-2, has resulted in a severe loss of human life worldwide. Although medications for COVID-19 and vaccination have enhanced the health outcomes of patients, there is increasing clinical evidence that COVID-19 may have long-term and persistent effects that can impact multiple organs and have severe clinical consequences called long COVID-19 or post-acute COVID-19 syndrome (PACS). Supervised machine learning (ML) techniques can be effective in the early detection of diseases. In this thesis, we applied five supervised machine learning classification models, Support Vector Machine (SVM), Decision Tree (DT), K-Nearest Neighbours (KNN), Random Forest (RF) and Multilayer Perceptron (MLP) using a real biomedical post-event COVID-19 dataset, along with a feature selection technique for early prediction of the impact of post-event COVID-19 or PACS in COVID-19 survivors. We then developed an adaptive Hybrid Soft Voting Ensemble Learning Model (HSVEM) on these classification models to improve their performance. The performance of the HSVEM was then compared to the base classification models to identify the best model performance, which is considered a solution for early identifying patients at high-risk post-acute COVID-19 syndrome based on various evaluation criteria such as accuracy, recall, precision, F1-score and area under the curve (AUC). The result of the proposed model demonstrates high performance with an accuracy of 91%, recall of 91%, precision of 90% and F1-score of 90%, which can be effectively applied in healthcare to assist in improving the quality of treatment and managing COVID-19 survivors appropriately.