04·Applications and limitations

Introduction

Unsupervised learning without labels offers enormous flexibility — we use it for customer segmentation, anomaly detection, document clustering, image compression, gene analysis, and community discovery in graphs. The price of that flexibility is the absence of ground truth: there is no "accuracy" to report. In this lesson you will see the most common real-world applications of k-means, hierarchical, and PCA, learn internal measures (silhouette, Davies-Bouldin, Calinski-Harabasz) and external ones (ARI, NMI), learn to assess clustering stability via bootstrap, and understand when unsupervised learning is simply NOT the right tool — because the data lacks structure, is too noisy, or the problem demands prediction rather than exploration.