Bernard Ženko (2007) Learning predictive clustering rules. PhD thesis.
The predictive clustering approach to rule learning presented in the thesis is based on ideas from two machine learning subareas, predictive modeling and clustering. Both areas are usually regarded as completely different tasks, however, there are also some similarities between the two areas. Predictive clustering approach builds on these similarities. It constructs clusters of examples that are similar to each other, but in general takes both the descriptive and the target variables into account, and associates a predictive model to each constructed cluster. Methods for predictive clustering enable us to construct models for predicting multiple target variables, which are normally simpler than the corresponding set of models, each predicting a single variable. To this day, predictive clustering has been restricted to decision tree methods. Our goal was to extend predictive clustering approach to methods for learning rules. The newly developed algorithm is empirically evaluated on several single and multiple target classification and regression problems. Performance of the new method compares favorably to existing methods. Comparison of single target and multiple target prediction models shows that multiple target models offer comparable performance and drastically lower complexity than the corresponding sets of single target models.
Actions (login required)