Interpretation and optimization of the \(k\)-means algorithm. (Q464720)

The paper proposes a new algorithm called the divided \(k\)-means (DKM), which is developed as a limit case of another recently proposed alternative to \(k\)-means, i.e. smoothed \(k\)-means algorithm (SMOKA). Many illustrative examples are provided along with the theoretical explanations. Experiments demonstrate the higher efficiency of the DKM as opposed to the classical algorithm in the determined locally optimal partitions and its better CPU time than the SMOKA competitor. DKM contrasts to the original \(k\)-means approach only when data points appear in the Voronoi diagram during the iterative progression and results can still be improved.