early cancer allert Used AI

Devlog – Skin Cancer Detection AI

Today I worked on training a deep learning model based on EfficientNetB4 to detect skin cancer using the ISIC dataset.

Key steps completed during this 1h30min session:

Preprocessed the data using augmentation (rotation, zoom, flipping).

Used transfer learning with EfficientNetB4, freezing all layers except BatchNormalization layers.

Ran the training script main_run.py from the Model Training folder.

Achieved ~82% validation accuracy after 10 epochs.

Plotted training/validation accuracy & loss curves and saved the model weights.

Next steps:

Fine-tune learning rate and increase training time.

Address slight overfitting by adding more dropout and data balancing.