Improved k-means Algorithm using Weight Estimation
Md Moshiur Rahman, Md Abdul Masud · 2021 International Conference on Automation, Control and Mechatronics for Industry 4.0 (ACMI) · 2021
Clustering is an unsupervised method where similar data points are belonged in the same group so that maximum similarity are obtained among them. The K-means is a widely used algorithm to perform clustering. This algorithm is used for various purposes because it is simple and efficient. One of its main limitations is that it first chooses initial cluster centers randomly. In this paper, we propose a method, namely, IK-meansWE (Improved k-means using weight estimation), which overcomes the main limitations of the algorithm. This algorithm uses weight estimation method to transform multi-dimensional data into one-dimensional data. The one dimensional data are used to make ascending order which are divide into k parts. Then k initial cluster centers are chosen from k different groups. These selected initial centers are used to perform clustering solution in k-means clustering process. We have used both artificially generated and real world data-sets for experiments. The results of clustering using IK-meansWE algorithm are better than several K-means type algorithms, which shows the enhancement of the performance of K-means algorithm.