Classification as Mining and Use of Labeled Itemsets

We investigate the relationship between association and classification mining. The main issue in association mining is the discovery of interesting patterns of the data, so called itemsets. We introduce the notion of labeled itemsets and derive the surprising result that classification techniques such as decision trees, Naïve Bayes, Bayesian networks and Nearest Neighbor classifiers can all be seen as the mining and use of sets of labeled itemsets. This unified representation of apparently different classification methods provides new insights and enables the development of classification techniques that combine features of the previously named ones. Our on-going work shows that there is great potential in pursuing this direction to develop more efficient and accurate classification techniques.

• Type: Conference
• Year: 1999
• Where: DMKD
• Authors: Dimitris Meretakis, Beat Wüthrich
• Keywords: Accurate Classification Techniques, Classification Mining, Classification Techniques, Data Mining, DMKD 1999