Figure 3
The vertical axis is labeled “True Positive Rate” and the horizontal axis “False Positive Rate,” both scaled from 0.0 to 1.0 in increments of 0.2. Four curves are shown in the legend: “all seq. model” (dashed blue line), “seq. 3-5-7-9 model” (dotted red), “seq. 6-7-8-9 model” (dashed dark blue line), and “seq. 3-4-5 model” (dotted yellow). A baseline (dashed pink line) runs from (0.0, 0.0) to (1.0, 1.0). The “seq. 6-7-8-9 model” curve demonstrates the best performance, followed closely by the “seq. 3-5-7-9 model” and the “all seq. model.” These three curves start at (0.0, 0.0) and rise concave downward to (1.0, 1.0), nearly overlapping. The “seq. 3-4-5 model” curve performs the worst but still exceeds the random baseline, rising almost linearly from (0.0, 0.0) to (1.0, 1.0) The confusion matrix is titled “V S L 6-7-8-9” with 2 rows and 2 columns. Columns represent “Predicted” and rows represent “Actual,” both with headers “bankruptcy” and “healthy.” A color scale on the right ranges from 390 (light blue) to 2350 (dark blue). Row entries are: Row 1: Actual bankruptcy: Predicted bankruptcy: 2328. Predicted healthy: 393. Row 2: Actual healthy: Predicted bankruptcy: 432. Predicted healthy: 2347.ROC curve comparison over the VSL models (on the left), and confusion matrix of the best VSL model on the test set