Expert derived automatically generated classification trees: an example from pediatric cardiology

Abstract

Classification trees provide an attractively transparent discrimination technique and may be derived either from expert opinion or from data analysis. The authors considered a real and complex problem concerning the diagnosis of babies with suspected congenital heart disease into one of 27 classes. A full loss matrix for all possible misclassifications was obtained from clinical assessments. A tree derived from expert opinion was compared with trees derived from analysis of 571 past cases both for the full problem and for a subset of 6 diseases. Automatic methods for tree creation had problems with rare diseases. Inclusion of 'costs of misclassification' feedback on the training dataset improved the performance of data derived trees though they were generally outperformed by the expert tree.<>