A comparison of classification methods as diagnostic system: A case study on skin lesions

Area of Science:

• Medical Informatics
• Computational Biology
• Machine Learning in Healthcare

Background:

• Existing skin lesion classification methods often use varied datasets, limiting direct comparison.
• A consistent dataset is crucial for evaluating and improving diagnostic system accuracy.

Purpose of the Study:

• To evaluate and compare different classification techniques for a skin lesion diagnostic system using a single, consistent dataset.
• To determine the most effective classification and feature selection methods for improved diagnostic accuracy.

Main Methods:

• K-Nearest Neighbor (KNN) with Sequential Scanning feature selection.
• KNN combined with Genetic Algorithm (GA) for feature selection.
• Artificial Neural Networks (ANN) with GA for feature selection.
• Adaptive Neuro-Fuzzy Inference System (ANFIS).

Main Results:

• KNN with GA-optimized feature selection achieved the highest accuracy at 94%.
• Adaptive Neuro-Fuzzy Inference System demonstrated strong performance with 92% accuracy.
• All tested methods provided satisfactory results, highlighting the potential of computational approaches.

Conclusions:

• Feature selection methods, particularly genetic algorithms, are key to enhancing classifier performance in skin lesion diagnosis.
• Diagnostic systems can serve as valuable decision support tools, augmenting the accuracy of human expert judgments.