Dimensionality Reduction and Classification of Dermatological Images using PCA and Machine Learning

Skin diseases pose grave diagnosis issues since they are highly similar among classes and have varied patterns over the various skin colors, particularly in Indian subjects. The current work proposes a mixed strategy using transfer learning-based feature extraction, dimensionality reduction, and traditional machine learning classification to effectively detect skin diseases. In an experiment conducted on a database of 9478 images for five dermatological classes, features were extracted from a pre-trained MobileNetV2 network. The statistical technique, Principal Component Analysis (PCA) was used to diminish feature dimensionality to facilitate effective visualization (3D PCA plots) and computational performance. Support Vector Machine (SVM) classifiers that used PCA-reduced features were highly accurate, with evident class separability illustrated in confusion matrices and performance metrics. The suggested framework emphasizes the promise of explainable PCA-based pipelines for skin disease analysis and presents a scalable solution for dermatological AI systems in resource-limited clinical environments.