Interpolation kernel machines are classifiers that fit the training data perfectly, achieving zero training error, and have been shown to generalize well to new data. They offer a strong alternative to other powerful classifiers such as support vector machines and should be considered when selecting classifiers. Dataset pruning is a technique that can enhance machine learning performance by removing redundant or less informative samples. In this study, we explore how training set pruning can improve the performance of interpolation kernel machines. Inspired by the curse of dimensionality, we utilized evolutionary algorithms (genetic algorithm and particle swarm optimization algorithm) for pruning the training set. Our experimental results reveal that pruning significantly enhances classification performance, demonstrating its effectiveness in optimizing interpolation kernel machines.