Diagnostic rules induced by an ensemble method for childhood leukemia

We introduce a new ensemble method based on decision tree to discover significant and diversified rules for subtype classification of childhood acute lymphoblastic leukemia, a heterogeneous disease with individual subtypes differing in their response to chemotherapy. Our approach simply uses each of top-ranked features as root node to build up different trees in the ensemble. Since these trees are all generated from original training samples, rules derived by our algorithm are true and reliable. This is a characteristic of our method contrast to state-of-the-art methods such as Bagging, Boosting and Random Forest which may produce false rules. Experimental results on a large gene expression profiling data set of childhood leukemia patients demonstrate that our proposed method is not only superior to other classifiers´ performance, but also can identify a small subset of genes for biomarker analysis.