CN2 algorithm

Abstract: We present a novel approach to machine learning, called ABML (argumentation based ML). This approach combines machine learning from examples with concepts from the field of argumentation. The idea is to provide expert's arguments, or reasons, for some of the learning examples. We require that the theory induced from the examples explains the examples in terms of the given reasons. Thus arguments constrain the combinatorial search among possible hypotheses, and also direct the search towards hypotheses that are more comprehensible in the light of expert's background knowledge. In this paper we implement ABCN2 as an extension of the CN2 rule learning algorithm, and analyze its advantages in comparison with the original CN2 algorithm.

Abstract: Most of the algorithms for extracting fuzzy classification rules generate conjunctive antecedents that use all the attributes of the system. Using this kind of antecedents, the number of rules grows exponentially in terms of the number of attributes of the system. This paper presents a new algorithm, FuzzyCN2, for extracting conjunctive fuzzy classification rules. This algorithm is a fuzzy version of the well known CN2 algorithm and produces an ordered list of fuzzy rules. FuzzyCN2 generates antecedents that may not include all the attributes of the system. These antecedents may cover a wide number of instances and, so, the number of extracted rules is smaller. The algorithm introduces the use of linguistic hedges as part of the selectors, thus producing more compact rules and reducing the number of generated rules.

Abstract: When performing primary reading on a newly taken radiographic examination, a radiologist often needs to reference relevant prior images of the same patient for confirmation of comparison purposes. To effectively support such prior image references, we proposed an intelligent patient image retrieval system, which includes a learning subsystem for including patient image extended knowledge, based on radiologists image reference behaviors. In this study, we extended CN2, a decision rule-based induction technique, to address challenging characteristics unique to our application, including missing/noisy data and multiple decision outcomes. Our experimental results showed that the overall learning performance of the proposed CN2-MD algorithm was satisfactory, exhibiting desirable noise tolerance, immunity to missing data, and robustness with limited training data.