Cluster Sort: A Novel Hybrid Approach to Efficient In-Place Sorting Using Data Clustering
Mohit Subramaniam, Tanya Tripathi, Omkumar Chandraumakantham · IEEE Access · 2025
This paper introduces a clustering-based in-place sorting algorithm, cluster sort. It is designed in such a way that it improves sorting efficiency by using data locality. It works in two phases: first, data elements are clustered based on the similarity in the values and then each of these clusters is sorted independently using comb or shell sort. This is a hybrid approach, and it helps minimize multiple comparisons within the clusters, which in turn improves performance, especially in large datasets with specific distributions. The traditional algorithms were selected due to their well-known efficiency and widespread use in sorting large datasets. This experiment of ours includes ordered, reverse-ordered, Gaussian, repeated values, same values, and uniform distributions. The results acquired from this experiment show that Cluster Sort (comb) outperforms Cluster Sort (shell) by 88% in ordered datasets and is 92.32% faster than Merge Sort. For reverse-ordered datasets, Cluster Sort (Comb) is 70.79% faster than Bucket Sort and 90.69% faster than Cluster Sort (Shell). In Gaussian distributions, Cluster Sort (Shell) improves by 25.88% over Bucket Sort and 74.41% over Merge Sort. In repeated-value datasets, although Quick Sort is faster, Cluster Sort (Shell) surpasses Bucket Sort by 13.64%. For uniform distributions, Cluster Sort (Shell) is 30.28% faster than Bucket Sort and 75% faster than Merge Sort. The reduction in unwanted comparisons through clustering is essentially what made Cluster Sort significantly outperform traditional algorithms in these datasets with inherent patterns, making Cluster Sort a coherent choice for practical applications.