An improvement of collaborative fuzzy clustering based on active semi-supervised learning

Semi-supervised clustering is a hybrid method of supervised and unsupervised clustering that has the advantages of both. The paper presents an improvement of the collaborative fuzzy clustering algorithm by using active semi-supervised learning. This method is implemented if there is very little labeled data and not enough to apply supervised learning algorithms. After each iteration, active semi-supervised learning is accomplished by adding samples to the labeled dataset. The initial data set has very few labeled samples. The proposed algorithm selects and adds more high-confidence data samples to the labeled data set. After each iteration, the labeled dataset is added with new data samples, which helps the clustering process be stable and achieve higher accuracy. Experiments on standard data sets from UCI Machine Learning Repository (UCI) and satellite image data show that the proposed method gives clustering results with high accuracy and stability compared to the pre-improvement algorithm.