Large Margin Rule‐Based Classifiers
Tibérius O. Bonates · Wiley Encyclopedia of Operations Research and Management Science · 2010
Abstract We introduce an optimization approach for the construction of large margin rule‐based classifiers. We base our description on the Logical Analysis of Data (LAD) methodology, but the same approach can be applied to different rule‐based classification algorithms. The novelty in our algorithm relies on the fact that it unifies the distinct tasks of pattern (rule) generation and creation of a so‐called discriminant function for classification. Moreover, the algorithm has a single parameter, thus, significantly reducing the necessity for parameter calibration when learning a new data set. We investigate how accurate the LAD classification models built with our algorithm are and how they compare to other machine learning models.