Aktif Öǧrenme Tabanli Ayrik Kodlama Yöntemi ile Veri Siniflandirilmasi

Dictionary learning is used in signal, image, audio and video processing applications to represent signals by a sparse set of atoms where sparse representations are managed for the problems of compression, denoising, feature extraction and data classification. In many machine learning applications, classifier ensembles are shown to be superior than their single classifier counterparts. On the other hand, possibility of access to huge amount of unlabeled data has been increased along with getting easy access to data. Active learning, which is proposed for this type of problems, is a learning method in which the most informative instances from the unlabeled data are chosen, then labeled by an oracle and after then added to the training set. In this study, proposed ensemble based active learning algorithm (Active-RDL) learns a dictionary for each class by use of random feature subspaces and in each iteration after labeling the most informative instances they are added to the training set. In the test phase, a class label is assigned by considering the majority of the outputs from ensemble classifiers. The final decision of the ensembles is given by the majority of the dictionary outputs. Active-RDL method is compared with ensemble based support vector machine (Active-RSVM) on ten different benchmarks from UCI repository. In the test results, proposed Active-RDL method outperforms Active-RSVM.