Detection of Skin Cancer Using KNN and Naive Bayes Algorithms

Abstract

Skin cancer is a disease with a low chance of survival, particularly melanoma, which is one of the deadliest. Furthermore, because of collectibles, low contrast, and relative presentation such as mole, scar, and so on, skin cancer progression is poorly organized from the skin lesion. Skin cancer is identified automatically utilizing lesion detection algorithms that have been refined for accuracy, efficacy, and performance. The suggested approach extracts features for early skin lesion detection using ABCD law, GLCM, and HOG join abstraction. Pre-processing is employed in the suggested work to increase skin aberration peculiarity and lucidity in order to eliminate artefacts, skin tone, hair, and so on. Geodesic Active Contour (GAC) was used for segmentation because it detects the irritated part autonomously, which is important for feature extraction. To extract equity, line, hiding, and expansiveness elements, the ABCD scoring approach was applied. HOG and GLCM were used to extract textural characteristics. The isolated characteristics are immediately fed to classifiers, which use various AI methodologies like as KNN and Nave Bayes classifiers to coordinate skin injury between compassionate and melanoma.

For this work, skin lesion photos from the International Skin Imaging Collaboration (ISIC) were retrieved, comprising 328 photos of large-hearted skin lesions and 672 photos of melanoma. The suggested system achieves 97.8 percent accuracy and 0.94 Area under Curve using Naive Bayes classifiers. Aside from that, using KNN, the Sensitivity was 86.2 percent and the Specificity was 85 percent