Fuzzy Preprocessing: Discovering Hidden Effects and Nonlinear Feature Relationships in Data Science and Machine Learning

In the feature selection stage, among the data preprocessing steps, the correlation matrix is used to identify dataset features that are linearly dependent on the output; however, the correlation matrix is incapable of detecting nonlinear effects or local variations. In this paper, a fuzzy logic-based method is proposed, in which the correlation matrix is first computed and features with low correlation to the output are identified. Then, using fuzzy logic, it is examined whether features with low correlation may have a significant effect on the output within a specific range. The results indicate that this method is not only effective for identifying features in datasets with nonlinear behavior or local variations, but also, owing to the inherent interpretability of fuzzy logic, it offers greater transparency and comprehensibility compared to conventional algorithmic methods.

کلیدواژه ها:

Feature Selection ، Fuzzy Logic ، Machine Learning ، Correlation Matrix