Using decision tree induction for discovering holes in data

Abstract

Existing research in machine learning and data mining has been focused on finding rules or regularities among the data cases. Recently, it was shown that those associations that are missing in data may also be interesting. These missing associations are the holes or empty regions. The existing algorithm for discovering holes has a number of shortcomings. It requires each hole to contain no data point, which is too restrictive for many real-life applications. It also has a very high complexity, and produces a huge number of holes. Additionally, the algorithm only works in a continuous space, and does not allow any discrete/nominal attribute. These drawbacks limit its applications. In this paper, we propose a novel approach to overcome these shortcomings. This approach transforms the holes-discovery problem into a supervised learning task, and then uses the decision tree induction technique for discovering holes in data.