Handling of Numeric Ranges with the Subdue System

Graph-based knowledge discovery has become a powerful tool in the machine learning and data mining areas. It provides a flexible and natural data representation to describe real world domains. In this research work we present a novel algorithm for graph-based approaches to deal with numerical attributes during the data processing phase implemented in the Subdue system. Our experimental results show that the use of numerical attributes increased classification accuracy in the Mutagenesis and PTC domains in 22% compared to the Subdue system when it does not use our numerical attributes handling approach. Our method also outperforms other author’s results for the same domains, around 7% for the Mutagenesis domain and around 17% for the PTC domain. Recently, artificial intelligence techniques have become tools to solve real world problems and the need to represent real world domains has increased. Many interesting real world domains have an inherently structured character for which we need to find data representations to effectively apply AI algorithms to them. Data domains maybe classified as flat, sequential, and structural according to their properties. Some of these domains contain important numeric attributes. Domains with continuous values are not appropriately manipulated by graph-based knowledge discovery systems, although they can be appropriately represented. To the best of our knowledge at the time of publishing this work a graph-based knowledge discovery algorithm that deals with continuous values attributes does not exist. A solution proposed in the literature to approach this problem is the use of discretization techniques as a pre-processing or post-processing step but not during the knowledge discovery phase. Adding this capacity to graph-based algorithms will allow us to improve the work with numeric attributes; in this way we will be able to improve the classification accuracy for the classification task and the patterns descriptive power.