A robust semi-supervised EM-based clustering algorithm with a reject option

In this paper we address the problem of semi-supervision in the framework of parametric clustering by using labeled and unlabeled data together Clustering algorithms can take advantage from few labeled instances in order to tune parameters, improve convergence and overcome local extrema due to bad initialization. We extend a robust parametric clustering algorithm able to manage outlier rejection to the semi-supervision approach. This is achieved by modifying the expectation-maximization algorithm. The proposed method shows good performance with respect to data structure discovering, even facing to outliers.