Multicollinearity-Based Attribute Selection in Multiobjective Fuzzy Genetics-Based Machine Learning for Fish Habitat Assessment

Abstract

Anthropogenic impacts such as urbanization and water pollution pose significant threats to freshwater ecosystems, leading to habitat degradation and biodiversity loss. Understanding species-habitat relationships is crucial for the conservation and restoration of these ecosystems. In this study, we applied Multiobjective Fuzzy Genetics-based Machine Learning (MoFGBML) to develop interpretable species distribution models for assessing the habitat suitability of five freshwater fish species. Since the number of attributes in the training data can significantly affects both the interpretability and classification accuracy of the fuzzy models, we implemented a multicollinearity-based attribute selection process using the variance inflation factor (VIF) and correlation matrix to identify and remove redundant attributes of the data. Our results demonstrate that the proposed attribute selection clearly reduces the model complexity at the small risk of the model accuracy. This provides a more transparent understanding of habitat suitability for target fish species.