A Novel Distance for Clustering to Support Mixed Data Attributes and Promote Data Reliability and Network Lifetime in Large Scale Wireless Sensor Networks

Abstract

Clustering based approaches in Wireless Sensor Networks helps in identifying the summarized data by exploiting the feature of data redundancy in sensor networks. Due to the inexpensive hardware used and unattended operation nature, nodes in the sensor networks are often prone to many failures malicious attacks and resource constraints and data collected in sensor networks are found to be unreliable. Moreover, the wide usages of sensor network in diverse application have put a constraint on sensor protocol to handle data of mixed types. To address the issues of energy minimization and data reliability, we propose a distributed agglomerative cluster based anomaly detection algorithm termed DACAD to detect the faulty readings based on kNN approach. Additionally, to support applications with mixed data attributes, we design a heterogeneous distance function, HOEM to handle both continuous and nominal attributes. In this paper we have evaluated the performance of proposed algorithm in terms of false alarm rate, false positive rate and detection rate. Our results demonstrate that the proposed distance achieves a comparable detection rate with low false alarm rate with a significant reduction in computation and communication over head and operates with both continuous and nominal data.