Model Performance Metrics for a Deep Learning-Ensemble System for Melanoma Recognition

Nine convolutional neural network models were evaluated for dermatoscopic differential diagnosis between malignant melanoma and other malignant skin lesions. The models, trained on the ISIC-2024 and HAM10000 datasets, achieved AUCs from 0.921 to 0.967, with an XGBoost ensemble model reaching an AUC of 0.988 on the test set. The dataset was created by Jinyan Jiang and last updated on 2026-05-15.

Use Cases

• Benchmarking deep learning model performance for skin lesion classification based on the nine CNN architectures mentioned.
• Training or validating ensemble learning methods for medical image analysis based on the described XGBoost integration.
• Researching feature importance in multi-model diagnostic systems based on the described use of raw scores as feature vectors.

Strengths

• Performance metrics are validated using multiple standard measures including accuracy, precision, recall, F1 Score, and AUC.
• The ensemble model achieved a high test AUC of 0.988, suggesting strong performance.
• The underlying data sources are two well-known dermatoscopic datasets, ISIC-2024 and HAM10000.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The dataset is small at 3.5 MB, which likely contains only summary results, not raw image data.

Provenance

Source

figshare

Collection Method

Results from training nine CNN models and an XGBoost ensemble model on the ISIC-2024 and HAM10000 datasets.

Freshness

Last updated 2026-05-15 04:28:17; freshness should be verified.