Enrico Blanzieri, Anton Bryl
In this paper we evaluate an instance-based spam filter based on the SVM nearest neighbor (SVM-NN) classifier, which combines the ideas of SVM and k-nearest neighbor. To label a message the classifier first finds k nearest labeled messages, and then an SVM model is trained on these k samples and used to label the unknown sample. Here we present preliminary results of the comparison of SVM-NN with SVM and k-NN.
Subjects: 3.1 Case-Based Reasoning; 12. Machine Learning and Discovery
Submitted: Feb 10, 2007