Conventional k-fold cross-validation suffers from the problem of using each instance for training (k-1) times and testing once, resulting in redundancy and a disproportionate impact on the learning process due to multiple instances. In this paper, we present a novel method, Irredundant k-fold cross-validation, which ensures that each instance is used for both training and testing exactly once throughout the validation process. This ensures balanced dataset utilization, mitigates overfitting due to instance repetition, and enables more distinct differences in model analysis. Experimental results demonstrate that this method maintains hierarchical and model-independent performance across diverse datasets, while providing lower variance estimates and significantly reducing overall computational costs due to non-overlapping training partitions.
