Positive Unlabeled Learning with Applications in Computational Biology

With the increasing amount of stored information, the use of data mining techniques to extract useful knowledge from them has become a key issue in many domains. Classifier induction algorithms are useful tools for this purpose as they allow us to summarise the information contained in the datasets into classification functions that can be used to make predictions on new data. One of the applications of classifier induction algorithms is the information retrieval. The classical approaches to this problem require the use of positive examples (objects of the kind we want to recover) and negative examples (non-positive ones), but negative examples are not always available. Learning from positive and unlablled examples is the topic of this book. The contributions presented in this work cover model induction, model averaging and feature subset selection algorithms and the evaluation of classifiers in absence of negative examples. In the applied part of the work presented in this book some of the proposed algorithms are used to solve two computational biology problems, the identification of genes associated with hereditary diseases and with cancer.