Аннотация
The Minimum-Sum-of-Squares (MSS) criterion is one of the most widely used formulations for partitioning points in Euclidean space. Although the classical k -Means algorithm is efficient, its iterative nature and sensitivity to initialization limit its scalability on massive datasets. This paper introduces a novel and scalable heuristic, the One-Pass k-Means (OPKM), which follows the Less Is More (LIMA) philosophy. OPKM processes the data sequentially in fixed-size batches, ensuring that each data point is visited exactly once, while maintaining weighted centroids that preserve historical information and mitigate early-chunk bias. An extensive experimental evaluation on five real-world datasets, including high-dimensional and large-scale cases with up to 14 million samples, compares OPKM against MiniBatch k -Means, Online k -Means, and a one-pass MiniBatch variant (MB-OnePass). In addition, synthetic stress tests are conducted to assess robustness under challenging streaming conditions such as ordered arrival, concept drift, imbalance, outliers, and high-dimensional noise. Across these settings, OPKM achieves competitive or improved clustering quality with substantially reduced computational effort. Owing to its simplicity, linear complexity, and single-pass design, OPKM provides a practical solution for large-scale and streaming clustering applications.