Multistrategy learner using Complimentary Algorithms

We present a hybrid learning algorithm called Learning Algorithm using SEarch Rings (LASER). LASER combines Naive Bayes and 1-Nearest neighbor learners in a multistrategic way. The Naive Bayes and the Nearest Neighbor algorithms seem to compliment each other in their search technique and their bias and present a interesting category of learning algorithms called complimentary learners. LASER uses kernel density estimates for evaluating locality information and determining which component algorithm to use. Empirical analysis shows that LASER was superior to the component algorithms. The main contribution of LASER was its robustness in handling class imbalance and feature irrelevance.