Extracting Refined Rules from Knowledge-Based Neural Networks
Machine Learning
CBMS'03 Proceedings of the 16th IEEE conference on Computer-based medical systems
Hi-index | 0.00 |
Transforming knowledge between connectionist and symbolic machine learning approaches is not a uniform task and there is no general recipe for performing it. Irrespective of the way the transformation of knowledge, the developed methods usually fall into two classes: methods with high rate of transformed knowledge is done, but with a lot of restrictions concerning the type of data in the training sets, training methods, and the size of data sets; and methods with moderate rate of transformed knowledge, but with less restrictions. Our main interest was to find or develop a technique that would possess the knowledge acquisition power of neural networks and explanation power of decision trees. That is why we developed a NN-DT Cascade method, which is capable of transforming a part of knowledge from the neural network into a decision tree.