Analisis K-Means Clustering pada Wireless Sensor Network Berdasarkan Residual Energy dan Packet Delivery Ratio

Abstract

Wireless Sensor Networks (WSNs) rely on nodes with constrained energy resources and efficient data transmission. This study uses the K-Means clustering algorithm to classify nodes based on residual energy and Packet Delivery Ratio (PDR) using a publicly available dataset containing 1,000 observations. PDR was derived from the number of packets sent and received, with values exceeding 100% limited to 100%. Both variables were then standardized using z-score normalization. The optimal number of clusters was evaluated using the Elbow method and silhouette analysis for values of k ranging from 1 to 10. The analysis identified k = 2 as the optimal choice, yielding the highest average silhouette score of 0.47015. The resulting clusters consisted of Cluster 1, comprising 262 observations (26.2%), and Cluster 2, containing 738 observations (73.8%). The mean PDR in Cluster 1 was lower at 43.21%, compared with 96.59% in Cluster 2. In contrast, the average residual energy was relatively comparable between the two clusters, measuring 1.464 and 1.447, respectively. These findings suggest that PDR was the primary variable distinguishing the two groups. The average silhouette scores for Cluster 1 and Cluster 2 were 0.399 and 0.495, respectively, indicating a moderately distinct clustering structure, although the separation was not entirely strong. The findings are exploratory and do not imply causal relationships. Furthermore, the results should be interpreted cautiously because 578 observations with PDR values exceeding 100% were capped at 100%.