Improved PSO-based algorithm for outlier detection
Dongyi Ye · Journal of Computer Applications · 2012
A new outlier detection method based on Particle Swarm Optimization(PSO) was recently proposed by Mohemmed,et al.(MOHEMMED A,ZHANG M,BROWNE W.Particle swarm optimisation for outlier detection [C]// GECCO'10: Proceedings of the 12th Annual Conference on Genetic and Evolutionary Computation.Oregon,Portland: ACM,2010:83-84).There exists an unreasonable phenomenon that its way of defining the fitness function does not necessarily ensure a good match with outlying degree of an object.A new fitness function by weakening the penalty on unreasonable radiuses was proposed so that the deviation between a particle's fitness and outlying degree of the corresponding data object was narrowed.The algorithm searched for an approximate optimal solution,and the radius was then determined to compute the outlying degree of each object.The experimental results on several UCI datasets show the superiority of the proposed outlier detection method with the new fitness function over the original one and the LOF algorithm.The study shows that a reasonable definition of fitness function contributes to the improvement in quality of outlier detection.