Supervised learning in presence of outliers, label noise and unobserved classes

Three important issues are often encountered in Supervised Classifica- tion: class-memberships are unreliable for some training units (Label Noise), a pro- portion of observations might depart from the bulk of the data structure (Outliers) and groups represented in the test set may have not been encountered earlier in the learn- ing phase (Unobserved Classes). The present work introduces a Robust and Adaptive Eigenvalue-Decomposition Discriminant Analysis (RAEDDA) capable of handling situations in which one or more of the afore described problems occur. Transductive and inductive robust EM-based procedures are proposed for parameter estimation and experiments on real data, artificially adulterated, are provided to underline the benefits of the proposed method.