Fig. (3).

Training and testing of a decision tree; (A) Decision tree built using 3 genes A, B and C for 6 poor responders (patients in orange) and 4 good responders (patients in green). At a single expression threshold, entropy gain for each gene is calculated and gene with highest entropy gain is selected as a node for splitting. For example, entropy gain for each gene is calculated (Table, left) and gene B is selected as it has the highest entropy gain (highlighted in red) for the root node. The root node is denoted by a diamond and the intermediate nodes are denoted by a circle. At threshold θ_b, gene B splits (black arrows) the training set such that 6 patients have gene B expression less than or equal to threshold θ_b and the 4 patients have expression above threshold θ_b. All 4 patients with gene B expression less than or equal to threshold θ_b belong to the poor responder category and, therefore, are represented in leaf node (light blue rectangle). (B) The predictive ability of the decision tree model is evaluated using a test set. (A higher resolution / colour version of this figure is available in the electronic copy of the article).