Biometric fingerprint recognition systems usually face the problem regarding how to deal with the high-dimensionality of features for processing and the potential performance degradation. In this study, we propose to apply the Principal Component Analysis (PCA) to Histogram of Oriented Gradients (HOG) feature for fingerprint classification with K-Nearest Neighbors (KNN) classifier. PCA has been already used in several biometry systems, but how it is combined together with HOG, and specially the classifier parameters influence, mainly in fingerprint recognition, has not yet been explored in depth. The work presented in this paper bridges this gap through a systematic analysis on two fingerprint datasets recorded under ideal and real-world conditions. We conducted a wide range of experiments on different image resolutions, PCA components and KNN neighborhoods. The results show that PCA with 150 components reduces the dimensionality at a proper scale maintaining the classification accuracy, whereas aggressive reduction via 400 components deteriorates it. In addition, we found that KNN with K = 3 had always produced better results than K > 3, suggesting that it must be necessary to carefully adjust parameters of the classifiers in addition to the reduction of features. Such observations lead to the advancement of efficient, scalable and robust fingerprint recognition systems and suggest practical strategies on how to optimize the biometric classification pipeline.