A Review on Analysis of K-Means Clustering Machine Learning Algorithm based on Unsupervised Learning
Purpose: This study revisits the K-means clustering algorithm as a core technique of unsupervised machine learning and explains, in plain terms, how it groups unlabeled data into meaningful clusters.
Methodology: The paper combines a narrative review of studies published mostly between 2021 and 2026 with a manual, step-by-step computation of K-means on a small one-dimensional dataset of nine values, using Euclidean distance and K=2. The final partition is then checked with the within-cluster sum of squares and the silhouette coefficient for K values from one to four.
Results: The algorithm stabilized after three reassignment rounds and produced the clusters {2, 3, 4, 10, 11, 12} and {20, 25, 30}, with centroids of 7 and 25. The within-cluster sum of squares fell from 348.00 to 150.00, and the mean silhouette coefficient reached 0.660, although K=3 gave a higher value of 0.711.
Conclusions: K-means is simple, fast, and easy to interpret, yet its output depends strongly on the chosen number of clusters and on the starting centroids.
Limitations: The worked example relies on a small one-dimensional dataset and does not test high-dimensional, noisy, or streaming data.
Contributions: The paper offers a transparent teaching case that links the classic procedure to current validation practice and to recent applied studies.