Melanoma Detection Using XGB Classifier Combined with Feature Extraction and K-Means SMOTE Techniques

Area of Science:

• Dermatology
• Computer Science
• Medical Imaging

Background:

• Melanoma is a severe skin cancer with high mortality if not treated early.
• Machine learning and deep learning show promise in computer-aided diagnosis of skin lesions.
• Existing methods face challenges in image feature extraction and handling imbalanced datasets.

Purpose of the Study:

• To develop an improved melanoma detection model.
• To address limitations in image feature extraction and data imbalance in skin lesion analysis.
• To enhance the accuracy and performance of computer-aided melanoma diagnosis.

Main Methods:

• Implemented transfer learning for automatic image feature extraction.
• Incorporated gender and age metadata into the model.
• Utilized an oversampling technique to manage imbalanced data.
• Compared various machine learning algorithms for optimal performance.

Main Results:

• The proposed improvement strategies yielded statistically significant performance enhancements.
• The ensemble model demonstrated superior performance compared to previous related models.
• The integration of metadata and advanced feature extraction techniques improved diagnostic accuracy.

Conclusions:

• The developed melanoma detection model offers significant improvements over existing approaches.
• The study highlights the effectiveness of transfer learning, metadata integration, and data balancing techniques.
• The proposed ensemble model represents a promising advancement in computer-aided melanoma diagnosis.