IEEE 17th Signal Processing and Communications Applications Conference, Antalya, Turkey, 9 - 11 April 2009, pp.650-653
In a classification problem, when there are multiple feature views and unlabeled examples, Co-training can be used to train two separate classifiers, label the unlabeled data points iteratively and then combine the resulting classifiers. Especially when the number of labeled examples is small due to expense or difficulty of obtaining labels, Co-training can improve classifier performance. In this paper, Co-MRMR algorithm which uses classifiers trained on different feature subsets for Co-training is used for audio music genre classification. The features are selected with MRMR (minimum redundancy maximum relevance)feature selection algorithm. Two different feature sets, obtained from Marsyas and Music Miner software are evaluated for Co-training. Experimental results show that Co-MRMR gives better results than the random subspace method for Co-training (RASCO) which was suggested by Wang et al. in 2008 and traditional Co-training algorithm.