LEAST SQUARE-BASED DIFFERENTIAL EVOLUTION ALGORITHM FOR n-DIMENSIONAL DATA CLUSTERING PROBLEM
The K-Means algorithm is commonly used for data clustering due to its simplicity and effective implementation. However, it has certain drawbacks. One of the main issues with K-Means is the random process involved in selecting the initial centroid, which can lead to varying results. To address this issue, many researchers have developed methodologies to enhance the performance of K-Means, aiming to achieve the optimal centroid. Data clustering is essentially a global optimization problem, making the objective function crucial. The traditional K-Means algorithm uses the Euclidean distance function as its objective function. However, this metric can sometimes result in non-separable clusters. To mitigate this effect, least squares terms are introduced. The objective function, dependent on the centroids, needs to be minimized using an appropriate method. This paper explores the impact of using metaheuristic approaches (such as differential evolution), a combination of differential evolution and K-Means, and the conventional K-Means algorithm on the formulated objective function to determine the optimal centroid. Computational performance data, such as silhouette score and CPU time, are collected during the computational phase. The results show that the combination of differential evolution and K-Means is more efficient based on these indicators. Additionally, numerical experiments are performed to solve the clustering problem.
Сведения
- Опубликовано
- 2025 · Т. 13, № 3, С. 36-49
- Язык
- Английский
- Цитирований
- 2
- ISSN
- 2306-6172 · 2308-9822
- Издатель
- L. N. Gumilyov Eurasian National University