2026-05-02T06:34:03Z https://meral.edu.mm/oai

oai:meral.edu.mm:recid/3117 2021-12-13T05:47:14Z 1582963413512:1596119372420 user-ytu

Systematic Selection of Initial Centroid for K-Means Document Clustering System Tin Thu Zar Win Moe Moe Aye <p>As the number of electronic documents generated<br> from&nbsp; worldwide&nbsp; source&nbsp; increases,&nbsp; it&nbsp; is&nbsp; hard&nbsp; to&nbsp; manually<br> organize,&nbsp; analyze&nbsp; and&nbsp; present&nbsp; these&nbsp; documents&nbsp; efficiently.<br> Document&nbsp; clustering&nbsp; is&nbsp; one&nbsp; of&nbsp; the&nbsp; traditionally&nbsp; data&nbsp; mining<br> techniques and an unsupervised learning paradigm. Fast and<br> high&nbsp; quality&nbsp; document&nbsp; clustering&nbsp; algorithms&nbsp; play&nbsp; an<br> important&nbsp; role&nbsp; in&nbsp; helping&nbsp; users&nbsp; to&nbsp; effectively&nbsp; navigate,<br> summarize and organize the information. K-Means algorithm<br> is&nbsp; the&nbsp; most&nbsp; commonly&nbsp; used&nbsp; partitioned&nbsp; clustering&nbsp; algorithm<br> because it can be easily implemented and is the most efficient<br> one in terms of execution times. However, the major problem<br> with&nbsp; this&nbsp; algorithm&nbsp; is&nbsp; that&nbsp; it&nbsp; is&nbsp; sensitive&nbsp; to&nbsp; the&nbsp; selection&nbsp; of<br> initial&nbsp; centroid&nbsp; and&nbsp; may&nbsp; converge&nbsp; to&nbsp; local&nbsp; optima.&nbsp; The<br> algorithm takes the initial cluster centre arbitrarily so it does<br> not always guarantee good clustering results. Different initial<br> cluster&nbsp; centres&nbsp; often&nbsp; lead&nbsp; to&nbsp; different&nbsp; clustering&nbsp; and&nbsp; thus<br> provide unstable clustering results. To overcome this problem,&nbsp; &nbsp;<br> Systematic Selection of Initial Centroid for K-Means (SSIC K-<br> Means)&nbsp; approach&nbsp; is&nbsp; proposed&nbsp; to&nbsp; improve&nbsp; the&nbsp; quality&nbsp; of<br> clustering&nbsp; in&nbsp; this&nbsp; paper.&nbsp; Unlike&nbsp; the&nbsp; traditional&nbsp; K-Means<br> clustering, the proposed SSIC K-Means method can generate<br> the&nbsp; most&nbsp; compact&nbsp; and&nbsp; stable&nbsp; clustering&nbsp; results&nbsp; based&nbsp; on<br> maximum distance initial centroids points instead of random<br> initial centroid points. In this paper, experimental results are<br> presented&nbsp; in&nbsp; F-measures&nbsp; using&nbsp; 20&nbsp; Newsgroup&nbsp; standard<br> datasets.&nbsp; The&nbsp; evaluations&nbsp; demonstrate&nbsp; that&nbsp; the&nbsp; proposed<br> solution outperforms the other initialization methods and can<br> be applied for other various standard datasets.</p> 2016-12-29 http://hdl.handle.net/20.500.12678/0000003117 https://meral.edu.mm/records/3117