Semi-Supervised Learning with the Integration of Fuzzy Clustering and Artificial Neural Network

Abstract

Supervised and unsupervised learning are different types of machine learning approaches that are used for pattern classification and clustering. Supervised learning finds the nearest matching by getting the knowledge from labeled training data whereas unsupervised learning does not acquire any knowledge of labeled data to produce output. In extensive research, it has been shown that to cluster huge amount of unlabeled data, the combination of both supervised and unsupervised learning provides more accuracy than the other conventional learning methods, which is the main concept of Semi-Supervised learning. In case of unsupervised learning, Fuzzy C-Means (FCM) is a well-known technique that uses membership matrix while doing clustering. The membership matrix can be used to identify uncertain points which are belonging to all the clusters with low or equal degree of membership. This fact motivated us to propose a Semi-Supervised technique with the integration of FCM and Artificial Neural Network (ANN). In this regard, ANN is trained twice with the certain amount of high membership and ground truth points. Thereafter, rest of the points is classified using such trained ANN. Here nine different well-known training functions are used in ANN, while top four are reported for different datasets. The result of the proposed method is compared with the other state-of-the-art methods for different artificial and real life datasets. For this purpose, two well-known metrics called Adjusted Rand Index and Minkowski Score are computed and reported quantitatively and visually. Moreover, the statistical significance of the results produced by the proposed method is justified using independent two-sample t-test.