GMM-Based Mean Approximation for Skin Disease Classification using Machine Learning

Skin infections are a matter of concern as they affect people's physical and mental health. Most of these conditions are due to pathogenic infections, which may reside inside the body and multiply over time, resulting in skin diseases. The symptoms appear as changes in skin texture, as lesions, or both. The analysis of texture features is the common way to identify skin diseases from images. A familiar way of extracting features is using filters on images. In this work, we use the Gray Level Co-occurrence Matrix(GLCM) for the feature extraction. While creating the dataset, each feature from the GLCM distribution is approximated using the Gaussian Matrix Model(GMM). For the experiment, the images from the Dermnet image library were downloaded, and six features namely Contrast, Correlation, Energy, Homogeneity, Dissimilarity, and Angular Second Moment(ASM) were extracted from these images. We tried classification using KNN, SVM, and Logistic Regression, in which KNN with Cosine Similarity metric gave better accuracy.