Construction of complex aggregates with random restart hill-climbing

Abstract

This paper presents the integration of complex aggregates in the construction of logical decision trees to address relational data mining tasks. Indeed, relational data mining implies aggregating properties of objects from secondary tables and complex aggregates are an expressive way to do so. However, the size of their search space is combinatorial and it cannot be explored exhaustively. This leads us to introduce a new algorithm to build relevant complex aggregate features. This algorithm uses random restart hill-climbing to build complex aggregation conditions. The algorithm shows good results on both artificial data and real-world data.