This script trains a deep learning model for skin disease classification on a dataset of categorized skin disease images. The model is built on MobileNetV2, an efficient and lightweight architecture ideal for resource-limited environments. Key features include data augmentation, mixed precision training, and multiple callbacks to enhance performance.

You can find the model and code here: Kaggle - Skin Disease Model

The dataset used is available at: Kaggle - Skin Diseases Image Dataset

1. Data Augmentation:

   • Utilizes ImageDataGenerator with augmentations (shear, zoom, flip, rotation, brightness adjustment) to improve generalization and reduce overfitting.
   • Includes a validation split of 20%.

2. MobileNetV2 Base Model:

   • MobileNetV2 is used for transfer learning with pretrained weights, omitting the top layer (include_top=False).
   • The base model is initially frozen for early training epochs.

3. Model Architecture:

   • Sequential architecture includes MobileNetV2 followed by a global average pooling layer, a dense layer with 128 units, dropout, and a softmax layer for class probabilities.

4. Training Configuration:

   • Optimizer: Adam with a learning rate of 1e-4.
   • Loss function: categorical_crossentropy.
   • Metrics: accuracy.

5. Callbacks:

   • EarlyStopping to halt training if val_loss shows no improvement.
   • ReduceLROnPlateau to lower the learning rate when val_loss stagnates.
   • ModelCheckpoint to save the best-performing model based on val_loss.

6. Mixed Precision Training:

   • mixed_float16 precision accelerates training by optimizing GPU memory usage on Kaggle.

• Adjust dataset paths to match your Kaggle setup.
• Ensure MobileNetV2 weights are available in the specified path.
• Running the script in a Kaggle notebook will:
  • Load and preprocess data.
  • Train the model with callbacks.
  • Save the model as skin_disease_model_mobilenet.h5 for later use.

Outputs include:

• Training and validation accuracy and loss per epoch.
• Saved model file skin_disease_model_mobilenet.h5.
• Intermediate checkpoints saved as best_model.keras.

• Modify batch_size and epochs as needed based on resources and training time.
• For higher accuracy, consider unfreezing additional layers for fine-tuning.