Figure 4.

Number of instances labeled as negative and positive: (a) without data balancing (original dataset); (b) with 80% used for training and validation; (c) with 20% used for testing; and (d) with 80% used for training after applying SMOTE.