An Optimal Random Forest Classifier for Diagnosing Covid-19 from X-ray and CTscan Images

Abstract

The unprecedented rate of spread and gravity of Covid-19 infection has played havoc worldwide. The spread of Covid-19 has pushed the nations’ medical infrastructures to their limits, yet no one could contain the infection and loss of lives. Therefore, accurate and efficient diagnosis of the infection is of central importance to battle out the Covid-19 disease. RT-PCR, the customary diagnostic tool for Covid-19, is a time-consuming process. The RT-PCR results may take two to three days. In the wake of the new Covid-19 variants, the RT-PCR test may also result in false negative cases. Machine learning algorithms have been successful for auto diagnosing Covid-19 from epidemiological and medical image data. Applying a machine learning algorithm for Covid-19 detection from X-rays and CT scans images combined with RT-PCR results can augment the rate of Covid19 diagnosis. Therefore, designing machine learning algorithms for the early Covid-19 diagnosis is the need of an hour. In this paper, we present Random Forest (RF) classifier for Covid-19 diagnosis. We use the resampling techniques to resolve the class imbalance in the data. Combining the RF classifier with its finetuned hyperparameters and resampling techniques to address the class imbalance yield promising results. Moreover, a Genetic Algorithm is employed to find an optimal hyperparameter configuration for the RF classifier. The statistical result of the approach indicates that accuracy, geometric mean and f-measure are 0.94, 0.93 and 0.92 respectively. The suggested method achieves higher sensitivity and comparable performance to deep and transfer learning approaches which are relatively computationally expensive but less comprehensible.