A Hybrid Method For Mining Wireless Sensor Networks Data

Recently data management and processing for Wireless Sensor Networks (WSN) has become a topic of active research in several fields of computer science, such as distributed systems, database systems, and data mining. The main aim of deploying the sensor network applications is to make real time decision. Since, highly resource constraints sensor nodes and huge volume of fast changing nature of sensor data makes it impossible to store entire data and develop time and space efficient algorithms for real time decision making. This challenge motivates the research community to explore novel data mining techniques dealing with extracting knowledge from a large continuous arriving data from WSN. Traditional data mining techniques are not directly applicable to WSN due to the nature of sensor data and special characteristics and limitations of the sensor networks.This dissertation provides an overview of how traditional data mining techniques and algorithms are revised and improved to achieve good performance in sensor network environment. A comprehensive survey of existing data mining techniques specially developed for wireless sensor network is presented. We presented the most representative features, benefits and design challenges of these approaches. Additionally it presents technique based taxonomy and comparative tables to be used as guideline to select a technique suitable for the application at hand.Based on the limitations of existing technique we proposed a hybrid data mining Method for WSN which uses online data processing at sensor node combines with online learning at central side. Node uses their processing abilities to locally carry out data processing and transmit only the required and partially processed data called Local Model. Since the nodes are resource constraint in term of memory, energy and bandwidth, so single-pass algorithms are applied for data processing as the data is arriving continuously and not available for next scan. Local models are distributed on entire network, which are integrated at Sink which is resource sufficient as compared to sensor nodes. As a result a Network Model is computed that is more abstract as compared to local model. The network model then integrated at central sever to get the Global Model for entire network view. We proposed a Distributed Data Extraction (DDE) method to extract data from sensor networks by applying rules based clustering and association rule mining techniques. Result shows that DDE is an efficient method for clustering in term of lifetime, number of rounds, and data loss rate. A significant amount of sensor readings sent from the sensors to the data processing point(s) may be lost, corrupted or may be due to any sensor fault. To address this problem DDE is able to estimate the missing values from available sensor reading instead of requesting the sensor node to resend lost reading. Results show our proposed approach exhibits the maximum data accuracy and efficient data extraction.An online clustering method Sensory Data Miner (SDM) is proposed to classify the real-time sensor network data. Hierarchical clustering together with incremental learning is proposed to develop an intelligent approach to analyze and classify data. By using the incremental learning the clustering model is adoptable to new usage patterns. For instance, as new usage patterns are observed, they are incrementally learned by the model and grouped into appropriate cluster. Application of three clustering techniques, Hierarchical, EM and SimpleK-Means are evaluated on real dataset collected from medical unit in order to investigate the appropriate method for establishing data pattern classification. By comparing the results it is observed that hierarchical clustering outperforms others in term of number of clusters, cluster size, accuracy, update time and change rate.A prototype has been developed based on proposed a hybrid Method for WSN. A medical ward is monitored and controlled by using wireless sensor technology. In this prototype DDE method is used to extract data from sensors nodes and then transmit towards sink. For classifying and analysis we used the SDM method along incremental learning which helps to control and monitor the incoming sensor data of medical ward. The results show considerable improvement in managing and controlling the system.