A New Unsupervised Validation Index Model Suitable for Energy-Efficient Clustering Techniques in VANET

Abstract

Clustering evaluation techniques are important to check the clustering algorithm quality. High cluster similarity help to reduce the distance between a node to node within the cluster, also good separation was more important to avoid overlapping clusters. The network performance will increase and the signal will be high. Many researchers proposed different validation indexes such as Davies–Bouldin, Dunn, and Silhouette indexes. These cluster validation indexes focus on the internal or external cluster similarity, and some of them deal with both cases. The employing of graph-based distance to non-spherical clusters and selection of reference points will not be effective all the time because the average distance between reference points and all nodes will be changed dynamically such as in the VANET application. To solve this problem a dynamic sample node should be selected or the similarity of all nodes should be checked. This paper proposes a new Minimum intra-distance and Maximum inter-distance Index (M2I) to improve these indexes. The proposed index checks the internal similarity and the external distance among all nodes from cluster to cluster to ensure that high separation will occur. M2I checks the similarity from node to node within the cluster and cluster to cluster. The proposed index will be an improvement of all high-rank indexes. The proposed index was applied in different scenarios (VANET and real datasets scenarios) and compared with other indexes. The index result shows that the proposed M2I outperforms the others. The M2I accuracy is 100% in the VANET scenario and 89% in the real datasets scenario.

Keywords

• Clustering analysis
• cluster index validation
• energy clustering algorithms
• K-means
• unsupervised learning
• VANET clustering