CNN-Based Image Classification (Real vs. Fake Images)

• Designed and trained 6 CNN models using TensorFlow and Keras to classify 140K images from Kaggle (70K real, 70K fake). • Achieved 94% training accuracy and 88% testing accuracy through hyperparameter tuning and regularization. • Applied dropout, batch normalization, and model variation to enhance performance and reduce overfitting

Dataset

• deepfake and real image

Model Files (hosted on Hugging Face)

• model_1.keras
• model_2.keras
• model_3.keras
• model_4.keras
• model_5.keras
• model_6.keras