10.22376/ijpbs.2019.10.1.p1-12 Volume 6 Issue 2 2015 (April - June) The design of an ensemble guarantees success, only when its base classifiers make both limited errors as well as high accuracy. We address the problem of achieving the possible enhancement of accuracy of the learning models for diabetes data set.In this paper we design an ensemble by four types of Meta level classifiers for this purpose. The base classifiers for feeding the ensemble we have proposed the pool of instance based learners uniformly and established the results of experiments by various size of neighborhood, selection of samples of training set and cross validation. Hence the introduction of such experiment we show the generation of an effective and diverse ensemble of NN classifiers. T.LAVANYA AND A.KUMARAVEL Ensemble, Unsupervised Learning, Bagging, Dagging, Multi boost, Ada boost, Instance based Learners, K-Nearest Neighbor Classifier, Classification accuracy. 1179-1186