Cross-Validation for K-Means Type Problem by Matthew Wu '25

Cross-Validation for K-Means Type Problem by Matthew Wu ’25, Wednesday May 14, 1:00 – 1:50pm, North Science Building 015, Wachenheim, Statistics Thesis Defense

Clustering algorithms such as K-Means often require the user to specify the number of clusters, K, as a hyperparameter. However, selecting an appropriate value for K —that is, a value close to the true number of clusters K^*—can be challenging, especially in the absence of prior knowledge about the dataset. This issue parallels model selection in supervised learning, where principled techniques like cross-validation are routinely used to choose the best model from a set of candidates. In unsupervised settings like cluster analysis, however, such model selection methods do not directly translate. Recent work has explored ways to adapt the cross-validation framework to clustering, but few studies have systematically evaluated these methods or established consensus on their effectiveness. This talk investigates and compares several cross-validation–based approaches for cluster analysis.