A popular approach to building rule models is association rule classification. However, these often produce larger models than most other rule learners, impeding the comprehensibility of the created classifiers. Also, these algorithms decouple discretization from model learning, often leading to a loss of predictive performance. This package presents an implementation of Quantitative Classification based on Associations (QCBA), which is a collection of postprocessing algorithms for rule models built over discretized data. The QCBA method improves the fit of the bins originally produced by discretization and performs additional pruning, resulting in models that are typically smaller and often more accurate. The qCBA package supports models created with multiple packages for rule-based classification available in CRAN, including arc, arulesCBA, rCBA and sbrl.
Tomáš Kliegr (Wed,) studied this question.