Priority Based Correlation Feature Set with Weighted Section Clustering for Melanoma Classification

Abstract

Detecting skin cancer at the right time becomes a thought-provoking problem. Diagnosis relies heavily on machine learning models. Skin cancer has become the most common kind of cancer among people of all ages. Early detection lowers mortality rates. This is achieved via a hybrid approach involving extraction of features for clustering and classification techniques. Skin lesion subclasses with the most relevant features, such as colour and texture, are allowed for categorization. In order to pick features from massive data sets that contain more unsuitable or repeated features, a methodology is devised. If not caught and treated early, melanoma, the worst form of skin cancer, can cause death. Only board-certified dermatologists can provide an accurate prediction. Due to a scarcity of qualified personnel, computer-assisted diagnosis has emerged as the method of choice. Selected feature based clustering using similar group of values and machine learning based classification technique for melanoma detection is proposed here. The system is put through its paces using a collection of dermoscopy images drawn from the normative database. In order to isolate the affected area, a feature similar based ranking clustering approach is used. Six separate color-texture feature extractors are used to pull the features from the clustered set. Disease categorization using machine learning is hindered by a lack of data and the curse of dimensionality. In this research feature dimensionality reduction model is applied that propose a Priority based Correlation Feature Set with Weighted Section Clustering (PbCFS- WSC) model. The proposed model performance is high in clustering and classification in melanoma classification when compared with the traditional models.