Optimizing American Sign Language Recognition via PCA-Driven Dimensionality Reduction and KNN Classification

This research work focuses on enhancing the recognition accuracy of American Sign Language (ASL) hand gestures using a machine learning pipeline that combines Principal Component Analysis (PCA) and the k-Nearest Neighbors (kNN) algorithm. A dataset containing over 34,000 grayscale images of ASL hand signs was used. Each image, originally represented by 784 pixel values, was preprocessed and reduced in dimensionality using PCA. The reduced features were then normalized to the [0, 1] range to make distance metrics more reliable for classification. Three versions of the kNN classifier—based on Manhattan, Euclidean, and Chebyshev distance metrics—were trained and evaluated using stratified 10-fold cross-validation. The number of PCA components was varied from 5 to 12 to examine how dimensionality affects performance. 🏆 Key Outcomes kNN with Manhattan distance consistently outperformed the other two metrics. The model achieved near-perfect classification with 12 PCA components: Accuracy: 0.997 AUC: 1.000 F1 Score: 0.997 Dimensionality reduction significantly improved performance while reducing computational complexity. This research work demonstrates the power of combining PCA with distance-based classifiers for high-dimensional image data and serves as a reference for scalable gesture recognition systems.