An Empirical Comparison of Outlier Detection Methods

Four outlier detection methods are compared using both publicly available smaller statistical datasets and real-life Knowledge Discovery in Databases (KDD) datasets [1]. The smaller datasets provide insight (via visualisations) into the relative strengths and weaknesses of the compared methods. The real-life large datasets test scalability and practicality of application. We are unaware of previous comparisons of outlier detection methods for data mining applications. A methodology for comparing outlier detection methods is developed and we provide performance benchmarks against which new outlier detection methods