A First Approach in the Class Noise Filtering Approaches for Fuzzy Subgroup Discovery

The presence of noise in data is a common problem that produces several negative consequences, and is an unavoidable problem, which affects the data collection and data preparation processes in Data Mining applications, where errors commonly occur. The performance of the models built under such circumstances will heavily depend on the quality of the training data. Hence, problems containing noise are complex problems and accurate solutions are often difficult to achieve without using specialized techniques. A particular supervised learning field as subgroup discovery has overlooked the analysis of noise and its impact in the description obtained. In this paper, the noise impact in subgroup discovery is analyzed in a complete experimental study, using recent filtering techniques for several class noise levels. Specifically, the analysis is performed through the FuGePSD algorithmwhich is a state-of-the-art SD algorithm based on genetic programming and fuzzy logic.