Describing Data Table with Best Decision

We propose a new way for describing data tables that is inspired by decision trees. Our goal is to summarize entire data table with one “average” object called best decision. The best decision is defined here as a decision that achieves the greatest value of a weight function. In our paper we first review computationally simple weight function for defining the best decision which does not account for the dependencies between the attributes. Then we define decision as a branch in a decision tree and introduce a weight function that takes those dependencies into account. As search-space for such decision grows factorially with the number of attributes, efficient pruning techniques are necessary. We define three pruning techniques that can be applied in combination. We present some empirical data to demonstrate the effectiveness of such techniques. Key-Words: data mining, decision tree, best decision, optimization, algorithms