Page 75 - Big Data Analytics for Intelligent Healthcare Management

68 CHAPTER 4 TRANSFER LEARNING AND SUPERVISED CLASSIFIER

Table 4.5 10-fold Cross Validation Results
Magnification Factor Wise 10-Fold Cross Validation Accuracy (%)
40× 100× 200× 400×
Feature Extractors Classifiers
ResNet-50 LR 94.17 94.41 94.09 92.03
SVM 90.72 90.32 90.06 89.28
K-NN 90.85 88.45 91.06 89.35
Inception V3 LR 92.60 92.18 91.80 88.73
SVM 92.54 90.04 90.55 89.00
K-NN 91.97 89.06 89.93 87.42
Inception ResNet V2 LR 92.98 92.13 92.86 88.80
SVM 94.04 92.30 94.22 89.42
K-NN 89.90 88.52 89.62 85.30
Xception LR 92.60 91.64 92.91 89.90
SVM 90.40 88.88 90.80 87.15
K-NN 89.71 88.88 88.26 85.44

Validation accuracy (%)
Xception+KNN
Xception+SVC
Xception+LR
IRV2+KNN
IRV2+SVC
IRV2+LR
IV3+KNN
IV3+SVM
IV3+LR
RN50+KNN
RN50+SVM
RN50+LR
87 88 89 90 91 92 93 94 95

FIG. 4.7
Validation accuracy graph for 40 .
4.6.2.2 Validation accuracy of 100×
Interpretation: On the 100 data, all of the combinations of feature extractors and classifiers gave
validation accuracy above 88% and the ResNet50 and LR classifier gave the best cross validation score
of 94.41% (Fig. 4.8).

4.6.2.3 Validation accuracy of 200×
Interpretation: On the 200 data, all of the combinations of feature extractors and classifiers gave a
validation accuracy above 88% and the Inception ResNet V2 with Support Vector Classifier gave the
best cross validation score of 94.22% (Fig. 4.9).