Statistical Box Graph Method and Determination of Outlier Values by Meta Heuristic Algorithms

Detection of outliers in the context of multivariate data is an important issue for data mining. Most of recommended methods for solving this problem are based on mahalanobis distance, euclidean distance, which is the most common distribution of observations or it is based on projection pursuit approach. Metaheuristic methods automatically optimize distance parameters in applications where it is used to detect outliers, instead of manually adjusting them. In this study, metaheuristic algorithms were used to detect outliers in the context of multivariate data, which is an important problem of data mining. Detection of outliers and, if necessary, removal from the data is important both to improve the quality of the original data and to reduce the effects of ambiguous values in the database information discovery process. Outlier values obtained from 6 different data sets are presented in tabular form using meta algorithms in which particle swarm algorithm, genetic algorithm and differential evolution algorithm.