Predictability-based collective class association rule mining

摘要: Associative classification is rule-based involving candidate rules as criteria of that provide both highly accurate and easily interpretable results to decision makers. The important phase associative rule evaluation consisting ranking pruning in which bad are removed improve performance. Existing association mining algorithms relied on frequency-based methods such support confidence, failing sound statistical or computational measures for evaluation, often suffer from many redundant rules. In this research we propose predictability-based collective class based cross-validation with a new step. We measure the prediction accuracy each inner steps. split training dataset into sets test then evaluate predictive From several experiments, show proposed algorithm outperforms some existing while maintaining large number useful classifier. Furthermore, by applying real-life healthcare dataset, demonstrate it practical has potential reveal patterns dataset.