Emerging technologies have metamorphosed the nature of surveillance and monitoring applications, but the sensory data collected using various gadgets is poorly synchronized and the analysis of such data remain changeable. Over the years, the need for security and surveillance systems has changed significantly due to the influence of various events and attacks. Anomaly detection systems based on various soft computing techniques like Genetic algorithms, neural networks and fuzzy logic exist in literature. Recently data mining and state transition analysis are becoming important components in identifying intrusions. From a data mining perspective, sensor network problems are characterized by a large number of variables (sensors), producing a continuous stream of data, in a dynamic environment. There are two approaches to process the data generated from the sensors namely Centralized approach and Distributed approach. This paper proposes semantic based intrusion detection system based on centralized approach in the application layer. In the proposed scheme, state transition analysis, pattern matching and data mining techniques are integrated to maximize the detection accuracy. Patterns and rules are formulated based on the events detected by the sensors in WSN. The sink receives information about the several events happening in the coverage area and correlates the streaming data (events) in spatial domain (several sensors) and time domain. The proposed scheme is validated with a real time experimental set up that consists of sensor nodes. Number of scenarios is tested during the experimental phase. Experimental results show that the proposed hybrid intrusion detection scheme identifies the intrusion accurately. © EuroJournals Publishing, Inc. 2012.