Information-theoretic partition-based models for interpretable machine learning
Abstract
In this dissertation, we study partition-based models that can be used both for interpretable predictive modeling and for understanding data via interpretable patterns. Specifically, we study probabilistic rule-based models for multi-class classification and histogram models for discretization, explanatory data analysis, and conditional mutual information estimation.
Probabilistic RulesClassification RulesThe Minimum Description Length PrincipleConditional Mutual Information EstimationExplanatory Data AnalysisMultivariate DiscretizationInterpretable Machine Learning
