Classifier fitness based on accuracy
Evolutionary Computation
| Hi-index | 0.00 |
In this paper, we introduce fault tolerance to XCS and propose a new XCS framework called XCS with Fault Tolerance (XCS/FT). As an important branch of learning classifier systems, XCS has been proven capable of evolving maximally accurate, maximally general problem solutions. However, in practice, it oftentimes generates a lot of rules, which lower the readability of the evolved classification model, and thus, people may not be able to get the desired knowledge or useful information out of the model. Inspired by the fault tolerance mechanism proposed in field of data mining, we devise a new XCS framework by integrating the concept and mechanism of fault tolerance into XCS in order to reduce the number of classification rules and therefore to improve the readability of the generated prediction model. A series of $N$-multiplexer experiments, including 6-bit, 11-bit, 20-bit, and 37-bit multiplexers, are conducted to examine whether XCS/FT can accomplish its goal of design. According to the experimental results, XCS/FT can offer the same level of prediction accuracy on the test problems as XCS can, while the prediction model evolved by XCS/FT consists of significantly fewer classification rules.