In this paper, we have presented a novel concept for constructing ensemble of classifiers. Here, we have considered a situation where data is available over the period of time. If enough remotely located data points are available at the classification system, the current system may not cope with newer data instances. In that case, existing settings or parameters of the classifiers need to be modified to act properly on newer instances. In this paper, we have presented a general technique for detecting enough remotely located data points arrived at the classifiers so that existing classification model can longer suitable for the new situation and proposed a change of settings to cope with the newer situation. We have performed detail analysis of our approach. Our approach has showed satisfactory results in dynamic environments.