Efficient and Fast Initialization Algorithm for K-means Clustering

Abstract

The famous K-means clustering algorithm is sensitive to the selection of the initial centroids and may converge to a local minimum of the criterion function value. A new algorithm for initialization of the K-means clustering algorithm is presented. The proposed initial starting centroids procedure allows the K-means algorithm to converge to a "better" local minimum. Our algorithm shows that refined initial starting centroids indeed lead to improved solutions. A framework for implementing and testing various clustering algorithms is presented and used for developing and evaluating the algorithm.

Image Extracting from Ancient Arab Documents with Complex Structures

“…To be able to classify the document pixels in three different classes, the idea was to apply the K-means clustering algorithm with K = 2 to the feature vectors of the different pixels for the first time to separate between the text and the graphic, then a second time for the separation between the pixels belonging to the images and the pixels belonging to other types of graphic. In both steps, the initialization of the cluster centers for K-means is performed using the ElAgha initialization algorithm [23].…”

Balancing effort and benefit of K-means clustering algorithms in Big Data realms

“…On their part, El Agha & Ashour [23] claim that the following initialization strategy yields improved results. For s = 1, …, d , and i = 1, …, n , let x i ( s ) be the s -th coordinate of point x i , , and ν s = min x i ( s ).…”

An improvement to the K-means algorithm oriented to big data

“…For determining the centroids, several methods have been proposed such as [4], [5], [6]; however, random selection is the most commonly used. • Classification: The distance of each object to each of the cluster centroids is calculated, and the object is assigned to the cluster whose object-to-centroid distance is the smallest.…”