Document Type: Original Article
Department of Computer Engineering, Faculty of Engineering, Golestan University, Gorgan, Iran.
Department of Electrical Engineering, Faculty of Engineering, Golestan University, Gorgan, Iran.
Outliers and outlier detection are among the most important concepts of data processing in different applications. While there are many methods for outlier detection, each detection problem needs to be solved with the method most suited to its unique characteristics and features. This paper first classifies different outlier detection methods used in different fields and applications to provide a better understanding, and then presents a new fuzzy method for outlier detection. The proposed method uses the fuzzy logic and the local density to assign a point to data instances, and then determines whether a piece of data is normal or outlier based on the value of resulted membership function. Evaluation of the proposed outlier detection algorithm with synthetic datasets demonstrates its good accuracy; moreover, evaluation of the performance in solving real datasets show that the proposed method outperforms the k-means and K-NN algorithms.