Skin Cancer Detection using ResNet, GoogleNet, and Ensemble Learning

This project focuses on the classification of skin cancer using deep learning models on the HAM10000 dermatoscopic image dataset.

ResNet Vs GoogleNet:

A comparative study between ResNet and GoogleNet was conducted, varying key parameters like epochs (5, 10), optimizers (Adam, SGD), batch sizes (16, 32), and a fixed learning rate (0.0001).

ResNet performed better than GoogleNet because of its deeper design and skip connections. These results show that CNNs are effective for classifying skin lesions and can help in early skin cancer diagnosis.

Ensemble Learning:

To further improve accuracy and stability, an ensemble of six CNNs was used:

1. ResNet
2. GoogleNet
3. VGG
4. DenseNet
5. EfficientNet
6. MobileNet

Voting strategies (soft/hard) were applied to combine predictions.

Visualizations & Metrics:

For each individual model (ResNet, GoogleNet) and the ensemble, the following were recorded and visualized:

1. Training & Testing Accuracy
2. Training & Testing Loss
3. Performance Plots
4. Image Classification Predictions