A Novel Framework for Skin Lesion Classification Using Hierarchical Attention Stacked Ensemble and Matthews Correlation Coefficient Weighted Averaging

By
Jubaer Ahamed Bhuiyan
Anwar Hossain Efat
Md. Shifaul Hasan
Faniyam Maria Mansia
April 1, 2026

Digital Health

Overview

This report presents a novel framework for skin lesion classification that leverages hierarchical attention and ensemble learning techniques. The approach addresses challenges such as class imbalance and model interpretability, aiming to enhance diagnostic accuracy in dermatology.

Background

Skin lesions can indicate a range of dermatological conditions, from benign to malignant. Accurate and timely identification is crucial for preventing the progression of serious diseases, particularly skin cancer. Traditional diagnostic methods can be limited by human error, highlighting the need for advanced technologies like artificial intelligence to improve detection and classification.

Data Highlights

No numerical data available in the provided source material.

Key Findings

The proposed framework utilizes hierarchical attention mechanisms to improve feature extraction from skin lesion images.
Ensemble learning strategies are employed to enhance predictive accuracy by addressing class imbalance.
Challenges in existing machine learning models include reliance on abundant data and limited interpretability.
Pre-prediction stacking is introduced to complement traditional ensemble techniques, improving performance on high-variance images.
Timely identification of skin lesions is essential for preventing severe health outcomes, particularly in cases of melanoma.

Clinical Implications

The integration of AI in skin lesion classification can significantly enhance diagnostic accuracy and reduce the risk of misdiagnosis. Clinicians should consider adopting these advanced methodologies to support timely and effective patient management.

Conclusion

The novel framework for skin lesion classification represents a significant advancement in dermatological diagnostics, addressing key challenges and improving the potential for early detection of skin cancers.