Table 1.
Literature review summary.
Table 2.
Samples in each sport category in unbalanced and balanced condition.
Fig 1.
A general architecture of federated learning model.
Fig 2.
Flowchart of genetic algorithm.
Fig 3.
(a) Sample architecture of the deep neural network, which consists of 4 layers including input and output (b) Representation of weights of the deep neural network presented in a part using vector (c) Chromosome for deep neural network presented in part an in binary.
Fig 4.
Crossover operation.
Fig 5.
Mutation operation.
Table 3.
Parameter used in genetic algorithm.
Fig 6.
Architecture of AlexNet.
Fig 7.
Architecture of VGG19.
Fig 8.
Architecture of efficientNetB3.
Table 4.
Parameter used in federated learning model.
Fig 9.
a) ResNet50 architecture; b) Stem block; c) Block1-Stage 1; d) Block2-Stage 1; e) FC Block.
Fig 10.
(a) Input image; (b) Activation values after the first convolution operation; (c) Activation values after the batch normalization operation; (d) Activation values at convolution layer 2; (e) Activation values after the max-pooling operation.
Table 5.
Time spent to train the various base architecture models in federated learning using the proposed methodology and global averaging.
Fig 11.
Loss vs. Accuracy with the VGG19 base model over a balanced data set: a) 2 clients, b) 4 clients, c) 6 clients, d) 9 clients, e) 10 clients.
Fig 12.
Loss vs. Accuracy with the VGG19 base model over an unbalanced data set: a) 2 clients, b) 4 clients, c) 6 clients, d) 9 clients, e) 10 clients.
Fig 13.
Loss vs. Accuracy with the ResNet50 base model over a balanced data set: a) 2 clients, b) 4 clients, c) 6 clients, d) 9 clients, e) 10 clients.
Fig 14.
Loss vs. Accuracy with the ResNet50 base model over an unbalanced data set: a) 2 clients, b) 4 clients, c) 6 clients, d) 9 clients, e) 10 clients.
Fig 15.
Loss vs. Accuracy with the AlexNet base model over a balanced data set a) 2 clients; b) 4 clients; c) 6 clients; d) 9 clients; e) 10 clients.
Fig 16.
Loss vs. Accuracy with the AlexNet base model over an unbalanced data set: a) 2 clients, b) 4 clients, c) 6 clients, d) 9 clients, e) 10 clients.
Fig 17.
Loss vs. Accuracy with the EfficientNetB3 base model over a balanced data set a) 2 clients, b) 4 clients, c) 6 clients, d) 9 clients, e) 10 clients, and an unbalanced dataset e) 2 clients, f) 4 clients, g) 6 clients, h) 9 clients, i) 10 clients.
Fig 18.
Loss vs. Accuracy with the EfficientNetB3 base model over an unbalanced data set a) 2 clients; b) 4 clients; c) 6 clients; d) 9 clients; e) 10 clients.
Table 6.
A tabular representation of Recall, Precision, F1-Score for unbalanced and balanced dataset for all used model.
Fig 19.
Comparison of accuracy for an unbalanced dataset using different deep learning models as the base model in federated learning.
a) AlexNet as the base model; b) EfficientNetB3 as the base model; c) ResNet50 as the base model; d) VGG19 as the base model.
Fig 20.
Comparison of accuracy for a balanced dataset using different deep learning models as the base model and federated learning.
a) EfficientNetB3 as the base model; b) ResNet50 as the base model; c) AlexNet as the base model; d) VGG19 as the base model.
Fig 21.
Comparison of accuracy after compression using GA and using EfficientNetB3 as the base model a) Unbalanced b) Balanced data set.
Fig 22.
AUC-ROC with area under curve for VGG19 base model: a) 2 clients; b) 4 clients; c) 6 clients; d) 9 clients; e) 10 clients.
Fig 23.
AUC-ROC with an area under curve for ResNet50 base model: a) 2 clients; b) 4 clients; c) 6 clients; d) 9 clients; e) 10 clients.
Fig 24.
AUC-ROC with area under curve for AlexNet base model: a) 2 clients; b) 4 clients; c) 6 clients; d) 9 clients; e) 10 clients.
Fig 25.
AUC-ROC with area under curve for EfficientNetB3 base model: a) 2 clients; b) 4 clients; c) 6 clients; d) 9 clients; e) 10 clients.
Table 7.
Accuracy, F1 Score, precision, recall for other datasets with proposed algorithm over balance and unbalanced dataset.
Fig 26.
Comparison of model size after and before pruning.
Fig 27.
Comparison of inference time after and before pruning.