Cross-validation is a model evaluation technique in which the available labeled data is partitioned into complementary subsets, with the model trained on one portion and tested on the held-out portion, repeated across multiple folds to produce a robust estimate of generalization performance. In WiFi/CSI sensing research, it is critical for assessing whether models trained under one set of conditions — such as a specific site, user group, or environment — can reliably generalize beyond the training distribution, directly addressing concerns about overfitting and deployment feasibility that are central to systems like CrossSense. Common variants include k-fold cross-validation, where data is divided into k equally sized folds with each serving once as the test set, and leave-one-out cross-validation, which is particularly relevant in sensing contexts where data is scarce or where subject-independent or site-independent evaluation is required.
Dictionary term