Planning for Data Mining Tool (PDM)

We present a tool, PDM (Planning for Data Mining), based on Automated Planning that helps users (non necessarily experts on data mining) to perform DM (Data Mining) tasks. The starting point is a definition of the DM task to be carried out and the output is a set of plans that are executed in a DM tool to obtain a set of models and statistics. Plans are data-mining knowledge flows, i.e. sequences of DM actions that should be executed over the initial datasets to obtain the final models. However, the number of feasible plans that solve the same DM task is huge making necessary to rank them by some criterion. In a first approach, the ranking is performed following some expert estimations on the desired mining-results of the DM actions. Afterwards, these estimations are improved using machine learning techniques. In order to define the DM task, we use emerging standards, such as PMML (Predictive Model Markup Language). PMML is the leading standard for statistical and DM models and supported by over 20 vendors and organizations. With PMML, it is straightforward to develop a model on one system using one application and deploy the model on another system using another application. The PMML file is automatically translated into a planning problem described in PDDL2.1. So, any state-of-the art planner can be used to generate a plan (or plans), i.e. the sequence of DM actions that should be executed over the initial dataset to obtain the final model. Each plan or knowledge flow is executed by a machine learning engine. In our case, we employ one of the most used DM tools, WEKA (Witten and Frank 2005). In WEKA, knowledge flows are described as files with a specific format, KFML, and datasets are described as ARFF (AttributeRelation File Format) files. The results of the DM process can be evaluated, and new plans may be requested to the planning system.