Mohammad Vaghefi

Data collection related to the flow pattern has always been associated with outliers due to various reasons. Outlier Detection in flow pattern experiments is of high importance and results in a better and more accurate understanding of the flow pattern. In this study, six data mining methods have been used to identify the outliers in flow pattern experiments. The discussed methods include: Box Plot, Histograms, Linear Regression, k-Nearest Neighbors, Local Outlier Factor, k-Medoids Clustering, Multi-Layer Perceptron, and Self-Organizing Map. The main aim of this study is to detect the outliers in data collection in order to conduct flow pattern experiments using the data mining methods. These methods have been analyzed and compared with each other in a case study and their performance evaluated. The experimental outliers under investigation were emanated from flow pattern experiments around a spur dike located in a 90 degree bend using Vectrino velocimeter (ADV). The range of velocity measurement of this device is between ±0.01 m/s to ±4 m/s and measurement accuracy is 1 mm/s. Also, the frequency is set at 50 Hz. The comparisons of different outlier detection methods results demonstrated that the Box Plot and the Local Outlier Factor methods have the best performance.