To create a shortlist of suitable classifiers for further refining, we ran the classifiers on the same data to compare them.
We chose to test 6 different classifiers for a spread of different approaches to classifying.
We tested them both with and without SMOTE preprocessing.
The following images show a comparison of the kappa and accuracy scores of the six classifiers. Kappa can be seen as a single value measure of the amount of true and false positives and negatives. A kappa closer to 1 is better, a kappa below 0.4 can be seen as poor. These guidelines are relatively arbitrary.
NaiveBayes is an implementation of a probabilistic classifier.
IBk is an implementation of a lazy learning method.
JRip is an implementation of a rule based method.
J48 is an implementation of a decision tree based method
SMO and LibSVM are implementation of a support vector machine method, with LibSVM implementing SMO.