Instance-based ensemble learning algorithm with stacking framework

Nowadays the most active research in supervised learning includes an integration of several base classifiers into the combined classification system. Such systems are known under the names multiple classifiers, ensembles methods. This topic attracts an interest of machine learning researchers as multiple classifiers are often much more accurate than the component classifiers that make them up. In this paper, we proposed a Lazy Stacking approach to classification (LS), a stacking framework with lazy local learning for building a classifier ensemble learner. Stacking is an ensemble that uses different “type” of base classifiers for labeling new instance. So by using stacking along with lazy learners, we can provide the desire accuracy. To investigate LS's performance, we test LS against four rival algorithms on a large suite of 10 real-world benchmark numeric datasets. Empirical results confirm that LS can statistically significantly outperform alternative methods in terms of classification accuracy.