Outlier Detection using Evolutionary Computing
Gutha Jaya Krishna, Vadlamani Ravi · 2016
In this paper, we proposed Harmony Search and Differential Evolution based outlier detection for medium dimensional numerical datasets. The sparsity coefficient is taken as the objective function for finding outliers in the data. The upper limit of the number of dimensions for a dataset is fixed using the threshold suggested by Chebyshev's inequality. A t-test is conducted on the optimal sparsity coefficient for both methods over 30 simulations. At a 1% level of significance, the t-test confirmed that the Harmony Search based method is statistically more significant than Differential Evolution based one for all four datasets. Both methods outperformed the previous approaches.