Two Novel Techniques for Finding Optimal K-value in K-means Clustering

  • Unique Paper ID: 161562
  • Volume: 10
  • Issue: 5
  • PageNo: 32-42
  • Abstract:
  • Two new techniques are proposed for determining optimal K-value in K-means clustering using decision tree classifier accuracy and its height. The first method is called Elbow Decision Tree Classifier (EDTC) created at elbow decision tree accuracy turning point and the second method is called decision tree classifier height (DTCH) determination at decision tree accuracy turning point. Standard UCI machine learning datasets are employed for experimentation purpose. Elbow turning point is a special K-value determined during decision tree accuracy starts to increase instead of usual accuracy decreasing. In EDTC, K-value at Elbow turning point is selected as the optimal K-value for K-means clustering. In the second proposed method (DTCH), decision tree height at the elbow turning point is taken as optimal K-value. The remarkable point is that Elbow K-value is approximately very close to the decision tree height. That is, approximately, equal optimal K-values in both the proposed methods is an indication that experiments are correct and consequently determined optimal K-values are also correct. Many standard UCI machine learning datasets are employed for experimentation purpose. Experiments results reveal that results are correct and optimal K-values determined in both the proposed methods are determined correctly.

Cite This Article

  • ISSN: 2349-6002
  • Volume: 10
  • Issue: 5
  • PageNo: 32-42

Two Novel Techniques for Finding Optimal K-value in K-means Clustering

Related Articles