MACHINE LEARNING ALGORITHMS FOR ERYTHEMATO-SQUAMOUS DISEASE CLASSIFICATION: FEATURE RANKINGS AND PERFORMANCE ANALYSIS

Erythemato-squamous diseases (ESDs), also known as erythrodermas, are a group of dermatological disorders characterized by both redness (erythema) and scaling (squamous) of the skin. These conditions can have various causes and implications. The implications of ESDs vary depending on the specific condition and its severity. While some may cause mild symptoms and have minimal impact on daily life, others can be chronic, recurrent, and significantly affect a person's physical and emotional well-being. Treatment options for these conditions may include topical medications, oral medications, phototherapy, and lifestyle modifications. In this paper, state of art machine learning (ML) algorithms is implemented for classification of ESD. To classify the disease a set of 11 clinical features and 23 histopathological features are considered. The performance of the ML classifiers is analyzed with individual sets of features and combination of both. Further, the performance of the ML classifiers is analyzed at different training rates to know the superior classifier for ESD classification. Furthermore, the study is extended to investigate the effectiveness of the Kruskal-Wallis algorithm in ranking the importance of features in the dataset used for disease classification. An investigation depicts that Ensemble and SVM classifiers outperformed the other ML classifiers in terms of accuracy and F1-score.