ADASYN: Oversampling Techniques for COVID-19 Prediction

In 2019, the highly contagious COVID-19 virus emerged and rapidly spread, causing a global health crisis characterized by inflammatory respiratory symptoms. The primary target of this disease is the respiratory system. Research has identified bats as the reservoir for the virus, but the intermediate host remains unknown. There is ongoing concern about various coronavirus variants, with the Omicron variant previously dominating globally and now attention shifting to a new variant, BA.2.86, informally known as Pirola, which is being closely monitored worldwide. However, the inherent class imbalance in COVID-19 datasets poses a significant obstacle to the effectiveness of these models. This paper proposes a novel approach combining outlier detection and Adaptive Synthetic Minority Oversampling Technique (ADASYNA) to address this issue. The Outlier-ADASYNA method identifies outliers within the minority class (COVID-19 cases) and synthesizes informative samples to rebalance the dataset. The proposed approach is evaluated using data from 5,400 patient records for both training and testing. The study employs ADASYN to effectively mitigate dataset imbalances and outstanding traditional oversampling techniques. This paper presents a comparative analysis of machine learning techniques, evaluating their performance using diverse metrics. The classification algorithm Random Forest attains the highest accuracy percentage of 90%, and RF Oversampling 92% surpassing numerous classification algorithms in predictive performance.