Decision trees using model ensemble-based nodes
Pattern Recognition
Using learning algorithm behavior to chart task space: The DICES distance
Intelligent Data Analysis
| Hi-index | 0.00 |
The problem of how to learn from examples has been studied thought the history of machine learning, and many successful learning algorithms have been developed. A problem that has received less attention is how to select which algorithm to use for a given task. The ability of a chosen algorithm to induce good generalization depends on how appropriate the model class underlying the algorithm is for the given task. We define an algorithm''s model class to be the representative language it uses to express a generalization of the examples. Supervised learning algorithms differ in their model class and how they search for a good generalization. Given this characterization, it is not surprising that some algorithms find better generalizations for some, but not all tasks. Therefore, in order to find the best generalization for each task, an automated learning system must search for the appropriate model class in addition to searching for the best generalization within the chosen class. This thesis proposal investigates the issues involved in the selection of the appropriate model class. The represented approach has two facets. Firstly, the approach combines different model classes in the form of a model combination decision tree,which allows the best representation to be found for each subconcept of the learning task. Secondly, which model class is the most appropriate is determined dynamically using a set of heuristic rules. Explicit in each rule are the conditions in which a particular model class is appropriate and if it is not, what should be done next. In addition to describing the approach, this proposal describes how the approach will be evaluated in order to demonstrate that it is both efficient and effective method for automatic model selection.