Detection and Classification of Bacterial Skin Infections Using K-Nearest Neighbors Algorithm

Abstract

Bacterial skin infections, including cellulitis and impetigo, pose significant health challenges requiring timely and accurate diagnosis for effective treatment. This research aims to develop an automated classification system for these infections using image processing and machine learning techniques. The study utilizes the Sobel method for image segmentation and Hu Moments for feature extraction. The classification is performed using the K-Nearest Neighbors (K-NN) algorithm with . The dataset, sourced from Kaggle, consists of imbalanced images of the two infection types. After pre-processing and feature extraction, the dataset is scaled to zero mean and unit variance. The model's performance is evaluated using cross-validation, yielding mean accuracy, precision, recall, F1-score, and ROC-AUC values of 65.95%, 65.18%, 65.95%, 63.06%, and 64.13%, respectively. Visualizations, including scatter plots, boxplots, histograms, correlation heatmaps, PCA, t-SNE, and UMAP, provide insights into the feature distributions and separability of classes. The results indicate that the combination of Sobel segmentation, Hu Moments, and K-NN can effectively classify bacterial skin infections. The study's contributions include demonstrating the applicability of these techniques to dermatological diagnostics and highlighting the potential for improved diagnostic accuracy and efficiency. However, the study acknowledges limitations such as data imbalance and variability in performance, suggesting the need for further research using advanced models like convolutional neural networks (CNNs) and enhanced data pre-processing techniques. These findings underscore the importance of machine learning in developing practical tools for clinical use, ultimately improving patient outcomes through early and accurate diagnosis.