Table 8 Comparative table of the different models used for comparison. Each of these models was evaluated in 5 classes: Any, scaling, PCA, and scaling + PCA, the hyperparameters used, and, finally, their accuracy is mentioned
Algorithms | Conditions on the dataset | Tuning Hyperparameters | Cross Validation [%] | Accuracy [%] |
|---|---|---|---|---|
DTC | Any | criterion = gini, min_samples_leaf =2, min_samples_split = 10 | 48.00 ± 2.0 | 48.74 |
Scaling | 48.00 ± 2.0 | 49.43 | ||
PCA | 55.00 ± 2.0 | 53.61 | ||
Scaling + PCA | 54.00 ± 1.0 | 54.91 | ||
MLP | Any | activation = relu, hidden_layer_sizes = (100,50), learning_rate = constant, solver = Adam | 30.00 ± 1.0 | 52.65 |
Scaling | 69.00 ± 1.0 | 72.39 | ||
PCA | 37.00 ± 1.0 | 59.09 | ||
Scaling + PCA | 69.00 ± 1.0 | 71.22 | ||
KNN | Any | algorithm = auto, leaf_size = 1, n_neighbors= 1, p = 2, weights = ‘uniform’ | 56.00 ± 1.0 | 54.26 |
Scaling | 56.00 ± 1.0 | 54.30 | ||
PCA | 57.00 ± 1.0 | 57.65 | ||
Scaling + PCA | 57.00 ± 1.0 | 57.61 | ||
ETC | Any | n_estimators = 300, random_state = 20, weights = ‘uniform’ | 72.00 ± 1.0 | 73.48 |
Scaling | 72.00± 1.0 | 73.48 | ||
PCA | 75.00 ± 1.0 | 75.87 | ||
Scaling + PCA | 76.00 ± 1.0 | 76.39 | ||
SVM | Any | C = 10, gama = ‘scale’, kernel = ‘rbf’ | 20.00 ± 1.0 | 19.96 |
Scaling | 64.00 ± 1.0 | 64.30 | ||
PCA | 19.00 ± 1.0 | 20.70 | ||
Scaling + PCA | 70.00 ± 1.0 | 71.09 | ||
RFC | Any | max_depth = None, min_samples_split = 2, n_estimators = 500, random_state = 40 | 73.00 ± 1.0 | 73.26 |
Scaling | 73.00 ± 1.0 | 73.22 | ||
PCA | 74.00 ± 1.0 | 73.48 | ||
Scaling + PCA | 74.00 ± 2.0 | 73.39 | ||
GB | Any | learning_rate = 0.1, max_depth = 7, n_estimators = 200, random_state = 10 | 69.00 ± 1.0 | 69.22 |
Scaling | 0.69 ± 1.0 | 69.26 | ||
PCA | 72.00 ± 1.0 | 72.61 | ||
Scaling + PCA | 71.00 ± 1.0 | 71.74 | ||
CNN+Fully | Any | optimizer=Adam(lr=0.001), epochs=500, batch_size=128 | 79.00 ± 2.0 | 83.83 |
Scaling | 85.00 ± 1.0 | 85.04 | ||
PCA | 55.00 ± 4.0 | 60.52 | ||
Scaling + PCA | 72.00 ± 3.0 | 72.04 | ||
CNN+Capsule-Net | Any | num_caps = 16, optimizer=Adam(lr=0.001), epochs=500, batch_size=128 | 79.00± 2.0 | 73.30 |
Scaling | 86.00 ± 1.0 | 87.13 | ||
PCA | 56.00 ± 3.0 | 56.65 | ||
Scaling + PCA | 72.00 ± 3.0 | 74.30 | ||
CNN+Tf | Any | NUM_HEADS = 16, NUM_LAYERS= 2, epochs=500, batch_size=128 | 79.00 ± 2.0 | 74.48 |
Scaling | 86.00 ± 2.0 | 88.34 | ||
PCA | 48.00 ± 1.0 | 49.35 | ||
Scaling + PCA | 72.00 ± 2.0 | 67.39 | ||
CNN+TF+Fully | Any | NUM_HEADS = 16, NUM_LAYERS= 1, epochs=500, batch_size=128 | 75.00 ± 1.0 | 76.74 |
Scaling | 85.00 ± 1.0 | 85.91 | ||
PCA | 57.00 ± 2.0 | 56.22 | ||
Scaling + PCA | 75.00 ± 1.0 | 75.00 | ||
CNN+Tf+Capsule-Net | Any | NUM_HEADS = 8, NUM_LAYERS = 1,num_caps = 16, epochs=500, batch_size=128 | 79.00 ± 2.0 | 80.83 |
Scaling | 86.00 ± 1.0 | 85.09 | ||
PCA | 53.00 ± 2.0 | 57.78 | ||
Scaling + PCA | 71.00 ± 2.0 | 72.35 |