Determining the Maximum Hyperbox Size in a Min-Max Fuzzy Classifier Using a Regression Model

Abstract: The paper proposes an algorithm for constructing fuzzy Min-Max classifier with adaptation of the maximum hyperbox size parameter using a regression model. The parameter-estimation model was developed using machine learning methods. To enable this, a set of 38 meta-features was proposed to character-ize dataset properties; these features were computed recurrently to support online learning. A computational experiment was conducted to construct classifiers using the proposed algorithm for solving cybersecurity problems such as spam detection, phishing-website detection, and network-attack detection. In the spam- and phishing-detection tasks, the proposed algorithm demonstrated a statistically significant improvement in accuracy compared to the Min-Max classifier that does not employ a regression model.

Keywords: automatic parameter selec-tion, cybersecurity, incremental learning, regression models, meta-features, fuzzy min-max classifier