Comparative Analysis of Binary Patterns Pyramid and Gabor Filters for Machine Learning-Based Nail Disease Detection

Abstract

Nail diseases, including Onychomycosis and Psoriasis, are common conditions that pose diagnostic challenges due to subtle clinical presentations and reliance on invasive methods. This study investigates the effectiveness of Binary Patterns Pyramid Filter (BPPF) and Gabor Filter (GF) as feature extraction techniques for machine learning-based nail disease detection, addressing three categories: Healthy, Onychomycosis, and Psoriasis. Using a dataset of 1,466 pre-labeled nail images from Kaggle, multiple classifiers AdaBoost, Random Forest, JRip, SVM, and Bayes Net were evaluated based on accuracy, precision, recall, ROC, PRC, Kappa, F Measure, and Matthews Correlation Coefficient. The results demonstrate that GF significantly outperforms BPPF across all classifiers, achieving a maximum accuracy of 90.94% with AdaBoost. To enhance novelty, a hybrid feature extraction approach combining BPPF and GF was proposed, leveraging the complementary strengths of both methods to improve classification performance. Furthermore, robust statistical validation using 10-fold cross-validation and paired t-tests confirms the significance of the observed differences. Feature importance analysis and computational cost comparisons provide additional insights into model efficiency and applicability. This study offers a comprehensive framework for noninvasive nail disease diagnosis, addressing both methodological gaps and practical challenges. The findings pave the way for real-time applications, including mobile and clinical diagnostic tools, highlighting the potential of machine learning in dermatological imaging.