Ensemble of Neural Networks Augmented with Noise Elimination

Developing a classifier (single) for determiningclass labels for unseen patterns in the life science domain is very common in the field of data mining and machine learning. However, as such data are very sensitive to noise/outliers, a classifier (single) in this context may not always be treated as a robust classification method. The literature has instead advocated combining many classifiers to increase overall accuracy, reduce the risk of classifier selection, and increase the robustness of the classifier. Therefore, in this work, we developed an ensemble of classifiers augmented with noise identification and a novel elimination method.This work is broadly twofold; for fold one, we use the density-based spatial clustering of applications with noise (DBSCAN) clustering technique to identify noise/outliers, which are subsequently eliminated by a novel method based on the high-sensitivity zone (HSZ) and keeping eye on the imbalance of class distribution. In the second step, the model is built using four base classifiers, such as multilayerperceptrons (MLPs) with back-propagation learning, radial basis function networks (RBFNs), extreme learning machines (ELMs), and functional link artificial neural networks (FLANNs). We conducted experimental studies on eight life science datasets collected from the UCI repository. The experimental study results support the claim that the suggested model has the potential to be more beneficial than classifiers (single)/ nonensemble classifiers.