A NEW APPROACH TO INSTANCE BASED LEARNING USING CLUSTERING

By combining the clustering idea with the law of gravity in physics, two new ideas are put forward to solve two important problems. One uses the number of the neighbor instances around one instance to describe its potential power and the other uses instance quality to predict a new instance class. Based on those two new ideas, a new approach to instance based learning using clustering is constructed. The results of the comparative experiment made on three complex data sets from the machine learning library show that this new approach outperforms many other instance based learning methods in aspects of the prediction power and the learning result memory requirement.