Applying sequential rules to protein localization prediction

In this paper we present a new classifier based on sequential classification rules for protein localization prediction. We also present three compact representations for encoding, in a concise form, the knowledge available in a classification rule set. Experiments run on the Gram-bacteria data set show that the classifier achieves both high prediction and good recall. Furthermore, since rules can be easily interpreted, biologists can understand classification results. To further improve classification performance, an SVM classifier is used to process data not covered by means of the sequential rule classifier.