Clinical features-based review of machine learning frameworks in predicting skin diseases

https://doi.org/10.36386/9rjn6t31

Skin disease is a medical illness that is often overlooked or underestimated among chronic ailments and deadly disorders. Meanwhile, the geographic dispersion of inhabitants who constantly experience excessive temperature being occasioned by hot weather increases the risk of skin disease. Though data mining and machine learning techniques play a vital role in grouping labelled or unlabelled data based on shared characteristics or similar attributes into predefined classes, the size and quality of data are germane to classification. In this study, commonly applied techniques for recent machine learning frameworks in the detection of skin disease, and their empirical comparison, were explored with a functional and conceptual review. The existing methods and models were examined for their specific usage and applications. The hidden but relevant patterns in clinical data were examined for proper fitting of the classification model, with relative parameters and the target variable to improve the sensitivity and accuracy of each classifier. Machine learning embedded functions in prominent data analytic tools allow automation of computational processes, like model building and validation, to maintain the sensitivity of clinical datasets. However, most of the selected machine learning techniques from previous studies were faced with rarely available real-time medical data, which could not be directly used for clinical analysis because a larger proportion of the medical data requires pre-processing before feature selection.