Post-processing data mining models for actionability

Data mining and machine learning algorithms are, in the most part, aimed at generating statistical models for decision making. These models are typically mathematical formulas or classification results on the test data. However, many of the output models do not themselves correspond to actions that can be executed. In this paper, we consider how to take the output of data mining algorithms as input, and produce collections of high-quality actions to perform in order to bring out the desired world states. This article gives an overview on two of our approaches in this actionable data mining framework, including an algorithm that extracts actions from decision trees and a system that generates high-utility association rules and an algorithm that can learn relational action models from frequent item sets for automatic planning. These two problems and solutions highlight our novel computational framework for actionable data mining.