Voting is an important Ensemble Learning technique. However, there has not been much discussion about leveraging the base classifiers’ consensus on unlabeled data to better inform the final prediction. My proposed method identifies the data points where the ensemble reaches consensus and where conflict arises in the unlabeled space. A meta weighted KNN model is trained upon this half-labeled set with the labels of the consensus and the conflict points marked as “Unknown,” which is treated as a new, additional class. The predictions of the meta model are expected to better inform the decision of the ensemble in the case of conflict. This research project aims to implement my proposed method and evaluate it on a range of benchmark datasets.
SOFTWARE ARCHITECTURE DIAGRAM: