Classification of haemoglobin typing chromatograms by neural networks and decision trees for thalassaemia screening

This article presents an application of a neural network and decision trees in thalassaemia screening. The aim is to classify thirteen classes of thalassaemia abnormality and one control class by inspecting the distribution of multiple types of haemoglobin in blood specimens, which are identified via high performance liquid chromatography (HPLC). C4.5 and random forests are the chosen architecture for decision tree implementation. For comparison, multilayer perceptrons are explored in classification via a neural network. The stratified 10-fold cross-validation results indicate that the best classification performance with overall accuracy of 97.2% (sensitivity = 97.2% and specificity = 99.8%) is achieved when C4.5 is used in conjunction with samples which have been pre-processed with input attribute discretisation and redundant attribute removal. Subsequently, C4.5 is applied to an additional sample set in a clinical trial which results in overall accuracy of 93.1% (sensitivity = 93.1% and specificity = 99.5%). These results suggest that a combination of C4.5 with haemoglobin typing analysis via HPLC may give rise to a guideline for further investigation of thalassaemia classification.