SVM-based Decision Tree for medical knowledge representation

Machine learning has become one of blooming research topics in recent years. Many applications can be found from integrating various techniques such as Chi-squared Automatic Interaction Detection (CHAID), Decision Tree, k-Nearest Neighbors (KNN), Recursive Partitioning and Regression Trees, and Support Vector Machines (SVM), to the obtrusive platforms that include the domains of healthcare, economics and agriculture. Researchers on healthcare domains have built effective systems to help clinicians alleviate diagnosis efforts. However, some models lacked flexibility to interpret the knowledge as if clinician's indulgement. To overcome such problems, SVM, one of the supervised learning algorithms with kernel radial basis function (RBF) as a nonlinear classification model, was exploited to classify and extract knowledge from medical data. The idea behind the proposed system was to classify the given data step by step by SVM. Incorrectly classified patterns will be fed to the succeeding stage to find a better split point in SVM. Split point was used to calculate information gain that can identify principal features from candidate attributes. Finally, knowledge-based decision trees were constructed from the ordered information gain to classify the unknown medical patterns. Experimental results from three different datasets verified that the proposed system was effective and feasible for the classification of medical databases.