Changelog

[Version 0.0.3] - Initial Release

• Initial release of the SkinVestigatorAI project.
• Data scraper, model training, and web application.
• Features the "Nano Model" for skin cancer detection.

Nano Model Details

• File size: 40 MB
• Trained on approximately 65000 images
• Quick training time

Here are the results from training the model:

• Training Epochs: 30
• Training steps per epoch: 1463
• Training loss: 0.2246
• Training accuracy: 46.41%
• Validation loss: 0.2526
• Validation accuracy: 46.15%
• Test dataset size: 6685 images
• Test loss: 0.2280
• Test accuracy: 46.38%

The model achieves a high accuracy of 91.38% on the test set. However, it's important to note that there is still some loss present in the model, which could potentially be improved upon in future iterations.

In future updates, the model's performance may be improved by incorporating additional training data, refining the model architecture, and optimizing hyperparameters. This would help enhance the overall efficiency and effectiveness of the skin cancer detection system.