Semi-supervised Clustering in Fuzzy Rule Generation

Inductive learning approaches traditionally categorized as supervised, which use labeled data sets, and unsupervised, which use unlabeled data sets in learning tasks. The great volume of available data and the cost involved in manual labeling has motivated the investigation of different solutions for machine learning tasks related to unlabeled data. The approach proposed here fits into this context: a semi-supervised clustering algorithm is applied to a partially labeled data set; the obtained results are used to automatically label the remaining data in the set; following, a supervised learning algorithm is used to generate fuzzy rules from the labeled data. The experiments show that this may be a promising solution for tasks that have encountered difficulties due to partially labeled data.