Application of Isolation Forest Technique for Outlier Detection in Water Quality Data
Jongeun Kim, No-Suk Park, Sangjin Yun, Seon-Ha Chae, Sukmin Yoon · 대한환경공학회지 · 2018
Although multivariate outlier detection techniques have been actively studied in various fields, domestic studies on the water quality management of waterworks are relatively inadequate and have been performed in a small range.In this study, water quality dataset were collected from G_water treatment plants in South Korea and classified by the statistical correlation.For the groups with significant correlations, we compared and analyzed the outlier detection performance by applying distance and isolation forest techniques.For the group with insignificant correlation, we analyzed the outlier detection performance according to the change of machine learning instance after applying Isolation Forest method.As a result, the distance-based and Isolation Forest methods were able to effectively search global and local outliers in the water quality dataset.Furthermore Isolation Forest method is analyzed to search outliers in a wider range than the distance-based method.In the Isolation Forest method, the change of the outlier search performance according to the change of the machine learning amount is small.