Abstract
Outliers significantly impact the reliability of the analyses informing the expert systems. Depending on the context in which data is collected, the traditional approaches of eliminating such outliers altogether or replacing them with pre-defined values could lead to the loss of genuine information. Machine learning techniques need to be tuned up to give suitable weights to unusual observations such as outliers. Robust methods are needed to minimise the influence of outliers on the reliability of such methods. In this study, we develop a fuzzy regression functions approach by the use of a noise cluster within fuzzy clustering algorithms that are potentially more robust against outlier contamination than the classical fuzzy c-means algorithm. The accuracy of the proposed approach is compared with those of traditional techniques such as artificial neural networks utilising various training algorithms, support vector machines using various kernel functions, and Fuzzy Regression Functions with a noise cluster and the fuzzy c-means algorithm. In total, 91 distinct implementations have been applied against 30 benchmark datasets with outliers present in both the independent and dependent features. The relative performances of these implementations are examined and the robustness of the proposed modified fuzzy regression functions framework is found superior to the considered alternatives.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have