Classifying Unseen Cases with Many Missing Values

Handling missing attribute values is an important issue for classiier learning, since missing attribute values in either training data or test (unseen) data aaect the prediction accuracy of learned classiiers. In many real KDD applications, attributes with missing values are very common. This paper studies the robust-ness of four recently developed committee learning techniques, including Boosting, Bagging, Sasc, and SascMB, relative to C4.5 for tolerating missing values in test data. Boosting is found to have a similar level of robustness to C4.5 for tolerating missing values in test data in terms of average error in a representative collection of natural domains under investigation. Bagging performs slightly better than Boosting, while Sasc and SascMB perform better than them in this regard, with SascMB performing best. Furthermore, we propose a novel voting weight scheme for the committee learning techniques. Although it is very simple, it can improve the robustness of all these four committee learning techniques for tolerating missing values in test data, especially when many missing values exist.