The computational complexity rate augments on training a classifier for a dataset with training samples as large as that of a nuclear power plant (NPP). Hence, there has been a constant focus to have classifiers with low training time. One of the methods to tackle such issue is using training dataset size reduction (TDR) techniques. TDR method leads to lowering memory requirements as well as training time. This paper delineates on the implementation of TDR method using Euclidean distance (TDR-ED) with the intention of crafting a hypothetical boundary which selects the necessary samples from the training dataset and dumps the rest. This approach is considered to be effective only if the accuracy of the classifier is not compromised even after reduction in the volume of the dataset. This is tested in this paper using some of the real world datasets to investigate the feasibility and efficiency of the TDR-ED approach using the combined effect of the classification accuracy and percentage reduction in the training dataset. Eventually, the feasibility of TDR-ED techniques for classification of some of the transients in a NPP is also inspected in this paper. © 2019 Elsevier Ltd