Linear machine weight adaptation in a genetic programming classifier that classifies medical data

While there has been a significant improvement in the overall decision tree classifier performance, not many methods focuses on the explicit treatment or measurement of sensitivity and specificity. Present methods generally pay less attention to the existence of misclassified input patterns and often fail to address the correction needed for error elimination or adjustment. This paper addresses the handling of the misclassification problem with the long term goal of improving the classifier accuracy in terms of sensitivity and specificity. The technique proposed is an oblique decision tree induction approach that relies on genetic programming (GP) and incorporates the linear machine decision tree algorithm through fitness evaluation. A robust GP fitness function handles generality and noise through weight adaptation during tree construction. By involving error correction each time the classifier is constructed, the proposed approach increases the classifier accuracy not only in terms of sensitivity but also specificity. The comparative evaluation of the proposed approach with selected classifier methods is presented in terms of accuracy, simplicity (size) and the construction time of the tree.