An outlier detection strategy for spatial free path-finding based on hierarchical ant colonies

Abstract

Outlier Detection (OD) is of great significance and widely used in various industries. It can not only find outlier objects to ensure industry safety, but also mine new knowledge. Considering the biological pattern of ant colony foraging behavior and inspired by swarm intelligence, this paper proposed an outlier detection strategy of spatial free path-finding for hierarchical ant colonies: OD Based on Path-finding of Hierarchical Ant Colonies (ODPHAC). Firstly, a new relationship measurement model between points is proposed to provide effective support for anomaly measurement. Then, two kinds of ant colonies with high and low levels are established, and different generation numbers are assigned to them. At the same time, different search ranges and path finding strategies are proposed for different levels of ant colonies to fully explore the data space and effectively save computing resources. Next, the representation method of data outlier-ness by bio pheromone is proposed, and an effective pheromone update operator is established to provide the necessary basis for ant colonies to find paths. Finally, the termination conditions of path-finding are put forward, and abnormal data results are produced. Under the support of the above-mentioned theory, the four models of Ant Generation, Range and Path-finding, Pheromone Update and Termination of Path Extension are established systematically. Through the sequential iteration of four models, the algorithm can explore the relevance of data space and assign anomaly degree to data objects. Experimental results show that ODPHAC is superior to 10 kinds of outlier detection algorithms in synthetic and real high-dimensional data sets. This paper is currently under review, but the relevant code can be found at GitHub.