A method for mining frequent patterns considering feature hierarchies

This article advances the author’s approach to solving data mining problems by integrating methods from explainable artificial intelligence and constraint programming theory. It proposes a method for mining frequent closed patterns that accounts for feature hierarchies. The approach is based on the construction a binary search tree and eliminates the need for a preliminary candidate generation stage. Feature hierarchy constraints are handled through specialized procedures that reduce the search space, thereby mitigating the effects of combinatorial explosion. In contrast to commonly used algorithms, the proposed method employs a depth-first rather than a breadth-first search tree traversal strategy. Its core component is a logical inference procedure that computes the closure of a given feature set. The method also supports the incorporation of additional constraints to further reduce the search space. Compared to existing approaches based on logical inference, it avoids redundant computations when determining closures across feature sets.