AAI_2025_Capstone_Chronicles_Combined

10 the hidden layer, 6 transformer layers, a learning rate of 10 -4 and a weight decay per epochof 10 -4 . The model achieved a 66% accuracy in the evaluation split in the ablation test. After selecting the final models, the CNN was trained for 5 epochs and the ViT for 10 epochs using the full training split and evaluated using the test split. Results Figure5 Model accuracy during training process

After training the best-performing model configurations on the full dataset, they were used to classify all the images labeled as Test. The CNN model achieved a total of 85.49% accuracy: Table1 Training test results of the CNN model Precision Recall F1Score Fake 0.82 0.92 0.86 Real 0.90 0.79 0.84

367