Assisting Data Mining through Automated Planning

Abstract

The induction of knowledge from a data set relies in the execution of multiple data mining actions: to apply filters to clean and select the data, to train different algorithms (clustering, classification, regression, association), to evaluate the results using different approaches (cross validation, statistical analysis), to visualize the results, etc. In a real data mining process, previous actions are executed several times, sometimes in a loop, until an accurate result is obtained. However, performing previous tasks requires a data mining engineer or expert which supervises the design and evaluate the whole process. The goal of this paper is to describe MOLE, an architecture to automatize the data mining process. The architecture assumes that the data mining process can be seen from a classical planning perspective, and hence, that classical planning tools can be used to design the process. MOLE is built and instantiated on the basis of i) standard languages to describe the data set and the data mining process; ii) available tools to design, execute and evaluate the data mining processes.