Fingerprint recognition remains a reliable biometric technique due to the uniqueness and permanence of fingerprint patterns. Nonetheless, challenges such as high feature dimensionality, sensitivity to rotation and partial impressions, and inconsistent acquisition conditions hinder classification accuracy and generalizability. This study proposes a robust fingerprint recognition framework that integrates Histogram of Oriented Gradients (HOG) for structural feature extraction and Principal Component Analysis (PCA) for dimensionality reduction. The resulting feature vectors are classified using six traditional machine learning algorithms: Support Vector Machine (SVM), Logistic Regression, Random Forest, Naïve Bayes, k-Nearest Neighbors (k-NN), and Decision Tree. Experiments were conducted on two fingerprint datasets: the public NUPT-FPV benchmark and a private dataset captured via ZKTECO devices under real-world conditions. The results demonstrate that SVM consistently achieved the highest accuracy across both datasets, followed by Logistic Regression and Random Forest. In contrast, Naïve Bayes and Decision Tree exhibited lower robustness and performance. The HOG–PCA pipeline effectively compresses fingerprint features while preserving discriminative information, establishing a reproducible benchmark for evaluating classical classifiers in fingerprint recognition systems. These findings offer valuable insights into classifier selection for biometric applications, balancing accuracy, computational efficiency, and generalization.