Figure 16
It displays two pairs of plots: two confusion matrices at the top and two learning curves at the bottom, comparing the performance of “Random Forest (R F)” and “Neural Network (N N)” models on an “Offices” dataset. Both matrix has three rows and three columns, with rows representing “True” values 0, 1, and 2 (top to bottom) and columns representing “Predicted” values 0, 1, and 2 (left to right). A color bar on the right indicates performance ranging from 0 percent (dark purple) to 100 percent (bright yellow), increasing in 25-percent increments through transitions from dark purple to green to yellow. The entries in the random forest matrix are as follows: Row 1: Column 1: 100. Column 2: 0. Column 3: 0. Row 2: Column 1: 4. Column 2: 92. Column 3: 4. Row 3: Column 1: 1. Column 2: 18. Column 3: 81. The entries in the neural network matrix are as follows: Row 1: Column 1: 83. Column 2: 17. Column 3: 0. Row 2: Column 1: 16. Column 2: 84. Column 3: 0. Row 3: Column 1: 1. Column 2: 15. Column 3: 84. The learning curve graph for “Random forest” is as follows: The horizontal axis is labeled “Training Set Size” and ranges from 50 to 400 in increments of 50 units. The vertical axis is labeled “Accuracy” and ranges from 0.5 to 0.9 in increments of 0.1 units. The graph shows two lines: one for “Training” and the other for “Validation.” The training accuracy starts high (near 0.88), increases slightly, and stabilizes at 0.88 accuracy after 150 set sizes. The validation accuracy starts low (below 0.5) and gradually increases to about 0.8, showing a moderate gap between training and validation accuracy. The learning curve graph for “Neural Network” is as follows: The horizontal axis is labeled “Training Set Size” and ranges from 100 to 600 in increments of 100 units. The vertical axis is labeled “Accuracy” and ranges from 0.5 to 0.9 in increments of 0.1 units. The graph shows two lines: one for “Training” and the other for “Validation.” The training accuracy starts high (near 0.62), and increases to about 0.84. The validation accuracy starts low (below 0.5) and gradually increases to about 0.88, indicating a moderate gap between the training and validation accuracies. Note: All numerical values are approximatedComparison of office models' confusion matrices (top) and accuracy vs training size curves (bottom). Source: Authors' own work