Page 78 - Big Data Analytics for Intelligent Healthcare Management

4.6 RESULT AND ANALYSIS 71

4.6.3 RESULT AND ANALYSIS OF TEST PERFORMANCE
4.6.3.1 Test performance on 40×
Interpretation: With ResNet50, LR gave the highest precision while Support Vector claimed the high-
est F1 Score and recall though they both have the highest accuracy because the LR model can correctly
classify the positive class better than the SVM model (Table 4.7, Fig. 4.11).
Interpretation: With InceptionV3, LR and SVM both have the highest recall value but the Support
Vector classifier had the maximum accuracy and precision, leading to the maximum f1score (Table 4.8,
Fig. 4.12).
Interpretation:
With Inception ResNet V2, SVM had the best accuracy, precision, recall, and f1score (Table 4.9,
Fig. 4.13).

Table 4.7 Result of ResNet-50 With LR, SVM, and K-NN on 40×

Feature Extractor Classifier Accuracy (%) Precision (%) Recall (%) F1-Score (%)
ResNet-50 LR 96.24 96.25 98.60 97.41
SVM 96.24 95.02 100 97.44
K-NN 93.23 92.71 97.80 95.19

102
100
98
96
94
92
90
88
Logistic regression Support vector K-NN
Accuracy Precision Recall F1score
FIG. 4.11
Performance of Resnet50 with three difference classifiers on 40 .

Table 4.8 Result of Inception V3 with LR, SVM, and K-NN on 40×

Feature Extractor Classifier Accuracy (%) Precision (%) Recall (%) F1-Score (%)
Inception V3 LR 94.49 94.30 98.25 96.23
SVM 94.74 94.61 98.25 96.40
K-NN 91.23 94.16 93.14 93.65