Selective sequencial integration on classification

To solve a classification problem we propose an algorithm that integrates sequencially Apriori association rule miner with C4.5 decision tree induction algorithm. This methodology uses constructive induction to extend primitive description language with interesting attributes. New attributes are selected frequent itemsets found using Apriori. Also we study the effect of different selection filters in algorithm accuracy and complexity. The experimental evaluation demonstrated that the proposed methodology generates high accurate and compact models. The best results were obtained using Pearson-correlation and heuristics from information theory as selection measures.