Beyond class imbalance and the sample size per class, this research paper studies the effect of knowledge-driven class generalization (KCG) on the accuracy of classical machine learning algorithms for mono-label classification. We apply our analysis on five classical machine learning models (Perception, Support Vectors, Random Forest, K-nearest neighbors, and Decision Tree) to classify a set of animal image files generated from Wikimedia Commons, a large-scale repository of free images. Thanks to our analysis, we found that the accuracy rates of mono-label classification models, mainly Support Vectors and K-nearest neighbors, are affected by KCG. The increasing or decreasing behavior of accuracy rates is driven by the settings of the classification categories and the generalization. The analysis of KCG should be useful to understand the limitations of classical machine-learning algorithms and to fuel debates about the improvement of classical models and supervised classification evaluation methods within the framework of Explainable Artificial Intelligence.