Combined clustering with classification in a semi-supervised context: An efficient data partitioning

Abstract

The studies on utilizing the output of unsupervised clustering techniques with a supervised classifier are pushing the concept of semi-supervised learning. Current ensemble models give the basic results and the consensus for the applied function. Very few attempts have been made to combine clustering methods with fundamental classifiers. The proposed method uses supervised clustering to partition the data into groups. The next step is to pairwise combine clusters from different groups to construct a number of training subsets. Each subgroup of the training set is given a unique base classifier, and he outputs of these base classifiers are consolidated through a specialized Consensus function. The weight given to a base classifier is based on how well it classifies the data. The experimental findings demonstrate that, compared to base classifiers and traditional ensemble classification methods, the proposed method ‘Ensemble of Clustering and Classification (ECC).’ gives a general performance upshift up to 10%.  Furthermore, it provides base classifiers with enhanced diversity and accuracy, thereby enhancing the data analysis process.

URN:NBN:sciencein.jist.2024.v12.824