Concept drift is the phenomenon of changes in the statistical characteristics of a data stream which effect the relationship between the inputs and the target variable leading to ineffectiveness of an ML model, and in extreme cases makes it obsolete. There are multiple types of concept drift and widely recognized methods are used to detect them but is widely accepted that there is no single method for all types of drift. Ensemble methods have been proposed but their base learners themselves are prone to drift. In this work statistical drift detection base learners are applied to three synthetically created datasets for each type of concept drift and then the results are compared. Furthermore, the drift detection methods are also applied to three real-world datasets. Afterwards, three stacking ensembles are created using different combinations of the drift detection methods and they are also applied to the datasets. Three Meta-learners are used and compared namely Logistic regression, XG Boost and Random Forest. The results show that all three outperform stand-alone drift detection algorithms on the simulated datasets and in case of real-world datasets, they are able to make correct predictions even when the stand alone base learners fail to do so. Furthermore, statistical base learners make sure that the methodology for drift detection isn’t circular i.e. prone to drift themselves as is the case with ML based drift detection algorithms.