Task Completed: Model Selection - CNN Architecture

During this week, focus was shifted to the selection of an appropriate deep learning model for deepfake image classification. After comparative analysis of various architectures, Convolutional Neural Networks (CNNs) were chosen due to their efficiency in spatial feature extraction from images.

Models Evaluated:

VGG16: Simple and effective but relatively heavy in terms of parameters.
ResNet50: Introduced residual connections to solve vanishing gradient problems in deep networks.
EfficientNetB0: Lightweight and high-performing; considered for later optimization.

Final Model Chosen:

Initially, ResNet50 was selected for transfer learning because of its balance between performance and computational efficiency. It allowed us to leverage pre-trained MobileNetV2 weights while fine-tuning on our specific deepfake dataset.