Concise and interpretable multi-label rule sets

Abstract Multi-label classification is becoming increasingly ubiquitous, but not much attention has been paid to interpretability. In this paper, we develop a multi-label classifier that can be represented as concise set of simple “if-then” rules, and thus, it offers better interpretability compared black-box models. Notably, our method able find small relevant patterns lead accurate classification, while existing rule-based classifiers are myopic wasteful in searching requiring large number rules achieve high accuracy. particular, formulate the problem choosing maximize target function, which considers only discrimination ability with respect labels, also diversity. Accounting for diversity helps avoid redundancy, control solution set. To tackle said maximization problem, propose 2-approximation algorithm, circumvents exponential-size search space using novel technique sample highly discriminative diverse rules. addition theoretical analysis, provide thorough experimental evaluation case study, indicate approach trade-off between predictive performance unmatched previous work.