Information Technology Reference
In-Depth Information
entire of the input rules at least once. After obtaining the training data set, a
rule evaluation model is obtained by using a learning algorithm.
In the prediction phase, a human expert receives predictions for the rest of a
new rules based on their objective index values.
Since rule evaluation models are used for predictions, we needed to choose
a learning algorithm with high accuracy similar to the current classification
problems.
6.3.2
Learning Algorithm Selection with Meta-learning
To enhance a classification task, people often use meta-learning algorithms.
One of the approaches integrates prepared base-level learning algorithms with
a meta-strategy such as voting, selecting and meta-level learning. We call this
approach “selective meta-learning”. In addition, we developed another meta-
learning scheme, which constructs a proper learning algorithm for a given dataset
using de-composed base-level learning algorithms. This approach is called “con-
structive meta-learning”.
In the field of meta-learning, there have been many studies on selective
meta-learning algorithms. There are two approaches for selective meta-learning
scheme. One includes bagging [5] and boosting [6], combining base-level
classifiers from multiple training data with different distributions. In these
Constructive meta-learning scheme
Base-level learning algorithms
Algorithm A
An adequate learnin algorighm
Algorithm B
Search for an adequate
learning algorithm,
re-composing leaning
algorithms
identified control structures
Algorithm C
identified similar methods,
having similar function
Given data set
Given data set
Selective meta-learning scheme
An combined classifier by
base-level and meta-level learning algorithms
Base-level learning algorithms
Algorithm A
Algorithm A
Algorithm A
Algorithm A
Algorithm B
Algorithm C
Algorithm A
Algorithm B
Algorithm C
Training base-level classifier
Training meta-level classifier
Meta-level learning algorithms/ strategy
Algorithm A
Algorithm B
Algorithm C
Algorithm A
Algorithm B
Algorithm C
Algorithm X
Algorithm Y
Voting
Fig. 6.2. Overview of constructive and selective meta-level processing scheme
 
Search WWH ::




Custom Search