Clustering Performance Analysis of K-Means and Expectation Maximization Algorithm of Machine Learning Algorithm

B., Ravindra Babu; Gebregziabher, Haftom; Abera, Kassaye Tilahun; Harika, Kunam

doi:10.22059/ijms.2026.421078.678814

Clustering Performance Analysis of K-Means and Expectation Maximization Algorithm of Machine Learning Algorithm

Articles in Press

Document Type : SI: MLFRM-2025

Authors

Ravindra Babu B. ¹

Haftom Gebregziabher ²

Kassaye Tilahun Abera ³

Kunam Harika ⁴

¹ ICT Department, Federal Technical and Vocational Training Institute (FTVTI), Addis Ababa, Ethiopia, 190310.

² Research Scholar in Knowledge Management, International School for Social and Business Studies (ISSBS), Mariborska cesta 7, 3000 Celje, Slovenia.

³ Faculty of ICT Department, Federal Technical and Vocational Training Institute (FTVTI), Addis Ababa, Ethiopia, 190310.

⁴ Sandhya Acqua PVT Ltd

10.22059/ijms.2026.421078.678814

Abstract

Clustering is an important unsupervised machine learning technique used to identify meaningful patterns and groups within unlabeled datasets. This study presents a comparative analysis of two widely used clustering algorithms, K-Means and Expectation Maximization (EM), for classifying Iris plant species. The publicly available Iris dataset, comprising measurements of sepal length, sepal width, petal length, and petal width, is used to evaluate the clustering performance of the two algorithms. K-Means is a partition-based clustering method that assigns observations to predefined clusters by minimizing within-cluster variation. In contrast, the EM algorithm employs a Gaussian Mixture Model (GMM) to estimate the probability distribution of observations and assign them to clusters based on probabilistic membership. The three Iris species—Setosa, Versicolor, and Virginica—are considered for the analysis. The clustering results are compared to assess the effectiveness of both approaches in identifying the underlying groups in the dataset. The findings indicate that the EM-GMM approach provides better clustering performance than K-Means for the Iris dataset, demonstrating its potential for capturing complex data distributions and improving cluster identification.

Keywords

Clustering

Data

Machine Learning

Iris, Unsupervised

Subjects