Machine learning-based monkeypox virus image prognosis with feature selection and advanced statistical loss function

Abstract

Recently, the monkeypox virus has gained paramount attention due to various complications entangled within it. These complications encompass pneumonia, eye problems, and secondary-skin infections. Current complications include swelling and sores within the rectum that would result in pain or complexity while urinating. Due to such complexities, it is crucial for monkeypox detection. Concurrently, with the evolvement of AI (Artificial Intelligence) based methods, existing works have tried to perform better detection of monkeypox and non-monkeypox. Nevertheless, these studies have been lagging in accuracy rate. As an enhancement, this study proposes RN-50-ZCA (Residual Network-50-Zero Phase Component Analysis) for feature extraction to attain enhanced classification performance. ZCA-whitening is utilized with RN-50, which assists in accurately identifying the features that agree with the image lesions. This approach incorporates data normalization and later linear transformation that has been considered to support lessening co-variance among the features. This also maintains the concrete variance. To fuse the features, PCA (Principal Component Analysis) is used. Finally, the research proposes MXGBoost (Modified eXtreme Gradient Boosting) based on statistical loss function for classifying monkeypox and non-monkeypox images (other viral samples, chickenpox samples, and smallpox samples) for acquiring effective prediction. Using MXGBoost with the loss function aids in extemporizing the prediction rate of the model by considering certain features of the issues being modelled. With such factors, the proposed loss function can support diminishing overfitting, thereby improvising the generalizability of the model. The performance of this study is assessed by comparison with three studies, and the analytical results exposed the better prediction rate of the proposed system.