A Hybrid PSO-MiLOF Approach for Outlier Detection in Streaming Data

Efficient outlier detection is important for the efficient performance and reliability of different types of telecommunication networks and applications, especially in wireless sensor network (WSN) scenarios. Still it is quite challenging to detect outliers in streaming data as the bulk of data to be examined is practically unbounded and can arrive at a high data rate. Most of the Local Outlier Factor (LOF) based algorithms suffer from large memory requirements, as well as high time complexity, hence they have limited use in practice for outlier detection on streaming data. To overcome the computationally intensive iterative training data stage of the LOF algorithms Swarm Intelligence (SI) based methods can be incorporated. Such hybrid outlier detection approaches can combine the properties of both LOF and SI methods. The utilization of Particle Swarm Optimization (PSO) based on bird flocks’ behavior are well known SI algorithms for solving optimization problems. This paper proposes a hybrid algorithm for outlier detection in streaming data based on PSO and Memory Efficient LOF. The major idea behind such an approach when solving the problem of finding the local outliers in streaming data is to meet the performance indicators of high detection accuracy, low computation time and limited memory usage.