Intrusion Detection System Based on KNN-MARS

The K-nearest neighbor (KNN) decision rule has been a ubiquitous classification tool with good scalability. In this paper, we propose a hybrid of KNN and MARS which deals naturally with the multi-class setting, has reasonable computational complexity both in training and at run time, and yields excellent results in practice. The basic idea is to find close neighbors to a query sample and train a local MARS that preserves the distance function on the collection of neighbors. A wide variety of distance functions are used and our experiments show performance on a number of benchmark data sets for IDS classification and object recognition. It draws a conclusion that KNN-MARS is a good method for multi-class setting.