Abstract: In order to solve the problem that the outlier detection of k-means algorithm is susceptible to fall into local optimality by the influence of the initial clustering center, an outlier detection based on the k-means algorithm of improving cuckoo search is proposed. Firstly, the adaptive strategy improvement is made to the discovery probability and Levy flight step size of the original cuckoo search algorithm, and the experimental simulation is carried out. Secondly, the convergence of the improved cuckoo search algorithm is discussed. Finally, the improved cuckoo search algorithm and the k-means outlier detection algorithm are fused into a new outlier detection algorithm: the outlier detection method based on the k-means algorithm of improved cuckoo search. Through the simulation experiments on UCI data sets, the results show that the proposed algorithm not only has obvious advantages in accuracy, but also improves the convergence speed on three data sets, which can effectively suppress the problem that the outlier detection of k-means algorithm is easy to fall into local optimality and shorten the running time.

Key words: outlier detection, k-means algorithm, cuckoo search algorithm, convergence