A FRAME WORK TO IMPROVE PERFORMANCE OF DISTRIBUTED DATA MINING USING RANKING AND MULTI-AGENT SYSTEMS

Distributed Data Mining (DDM) and Multi-Agent Systems (MAS) go hand in hand in collaborative problem solving in distributed environment. The data is geographically distributed at different sites. DDM with MAS can bestow the synergic effect in data intensive applications. Distributed clustering algorithms play a vital role in this context in many enterprise applications. Sensor networks are widely used in distributed environments for both military and civilian purposes. Sensor networks are resource constrained. In such networks designing and implementing a framework for supporting DDM with MAS is a challenging and non trivial problem. Moreoverprivacy preserving data mining is an essential requirement as nondisclosure of privacy is essential. However, the DDM with MAS for performing clustering in sensor networks throws many challenges. They include limited communication bandwidth, constrained computing resources and energy, need for fault-tolerance, and the asynchronous nature of the network. The existing research focused on either DDM with MAS or DDM for sensor and other networks. A comprehensive framework that accommodates DDM, MAS, and clustering, ranking, privacy preserving data mining in sensor networks is yet to be desired. In this research we propose and implement such framework to realize efficient clustering with privacy preserving in sensor networks with DDM and MAS. The name of the proposed framework is Agent Based Extended Distributed Data Clustering for Sensor Network (ABEDDCSN).The usage of agents in sensors in distributed environment provides promising applications in the real world.The DDM algorithms defined for clustering can enhance the capabilities of sensor networks as they can minelatent trends or patterns. Care of interpretation of the results of DDM in sensor networks can give rise to business intelligence for making expert decisions. In addition to this, the framework also supports the concept of multi-ranking along with PPDM to leverage the utility of the results of clustering.