Machine Learning-Based Melanoma Skin Cancer Identification Using Feature Fusion of Thepade's SBTC (TSBTC) and Niblack Thresholding

Skin Cancer is characterized by clinical science as an unusual development of skin cells. Skin cell malignant growth is one of the most dangerous illnesses that can be fatal if not identified in the early stages of development. One of the many types of malignant growths in skin cells is melanoma. Melanoma usually forms in melanocytes which are the pigment-producing cells of the skin. The traditional approach is time-consuming and relies on doctors' expertise and skin biopsy results. Subsequently, the paper centers around proposing a framework for the characterization of Melanoma skin disease images as Benign and Malignant. Global Feature Extraction using Thepade's SBTC (TSBTC) is employed for feature extraction ranging from 2-Ary to 10-Ary. Local features extraction based on Niblack Thresholding is also used to perform feature extraction for the Melanoma Dataset. Thepade's SBTC (TSBTC) and Niblack Thresholding are fused with global and local features extracted. The experimentation carried out using 1000 images categorized as 500 benign and 500 malignant images have shown better accuracy, sensitivity, and specificity. Fusion-based approach for Melanoma Cancer Identification gives promising results compared to individual global or local feature formation, proving the worth of feature fusion. A total of 12 Machine learning Classifiers are utilized, out of which 4 are ensembles.