SQL-based heuristics for selected KDD tasks over large data sets
Marcin Kowalski, Sebastian Stawicki · 2012
We investigate how to use the scripts with automatically generated fast-performing analytic SQL statements to speed up the KDD-related tasks of attribute selection and decision tree induction. We base our framework on the entity-attribute-value data model in order to seamlessly scale the required queries with respect to the amounts of attributes involved in the given task's specification. We note that the considered tasks can be heuristically handled using the same class of aggregation queries, where the most promising attributes and splits are searched by analyzing diversity of aggregated results grouped by decision. We also outline our plans with respect to creation of a large-scale framework for evaluating the proposed heuristics against real-world data.