RULES-IS Immune-network inspired machine learning algorithm
This paper proposes a novel algorithm for the extraction of rule sets. The extraction technique utilised is inspired by the immune system and its pattern recognition capabilities. The concept of an immune network is used which enables the algorithm to generate rule sets in an incremental manner. Such an immune network also enables the algorithm to learn from training examples that have not had classes specified. The algorithm developed is tested against several different example data sets and shows itself to be at least comparable to existing algorithms.
Bonjour, Samuel: Many thanks for your interest in our paper. I would argue that RULES-IS is more able to generalise by not trying to learn the training set perfectly (which is the case with RULES-3). This better generalisation is demonstrated by the higher accuracy on the unseen test data. You are correct about the need to compare with the more recent versions of RULES. We will leave this for a future occasion.
First of all, thank you for this interesting concept.
You mentioned that one of the main feature of the algorithm is that it "Helps prevent creation of rules that misclassify"
However, the result on Iris training set shows some misclassification. What is the reason for this? Is the algorithm able to deal with noisy example?
Finally, the results does appear good, it would be interesting to compare with more recent algorithms such as Rules5 and Rules6.
Samuel Bigot