COVID-19 is a contagious disease and the most extensive health disaster, with daily increases in infected people and fatalities. Coronavirus variants such as Omicron and Delta are spreading rapidly, which are more dangerous for people of all ages. Due to its prevalence in humans, the prediction and discovery of the Omicron variation of COVID-19 raised significant concerns among the medical community.The frequency and severity of chronic diseases are rising, and population aging results in an increased risk of COVID-19 related to hospitalization or mortality. Finding potentially high-risk patients is complex, and the diagnostic procedures for this COVID-19 variation have become more intricate. Researchers and experts are involved in establishing quicker and less expensive diagnostic approaches. Due to the high population growth and spread of disease, automatic disease diagnosis and prediction have become essential topics in the medical industry. The machine learning (ML) technique, such as an improved random forest model, is proposed to predict COVID-19 and its variants in this work.It is a two-phase process; the bootstrap resampling method is employed for obtaining the original samples and applying a decision tree voting approach for prediction. The obtained accuracy by the proposed RF model is compared with CNN [12] and ECNN-ERNN [13]. The proposed RF model archives 98.8% of accuracy, which defines its enhanced efficiency more than the others.