A novel melanoma prediction model for imbalanced data using optimized SqueezeNet by bald eagle search optimization.

Area of Science:

• Dermatology and Medical Imaging
• Artificial Intelligence in Healthcare
• Computational Biology

Background:

• Accurate skin lesion classification is vital for diagnosing skin conditions, including melanoma.
• Existing datasets, like ISIC 2020, present challenges due to severe class imbalance, hindering model performance.
• Traditional diagnostic methods require expert interpretation and can be time-consuming.

Purpose of the Study:

• To develop and evaluate a novel hybrid deep learning model for classifying skin lesions as normal or melanoma.
• To address the challenge of class imbalance in the ISIC 2020 dataset.
• To optimize a convolutional neural network architecture using a metaheuristic algorithm for improved diagnostic accuracy.

Main Methods:

• A hybrid model combining a convolutional neural network (SqueezeNet) with Bald Eagle Search (BES) optimization was proposed.
• Random over-sampling and data augmentation techniques were employed to mitigate class imbalance.
• The BES algorithm was utilized to fine-tune the hyperparameters of the SqueezeNet architecture.

Main Results:

• The proposed model achieved high performance metrics: 98.37% overall accuracy, 96.47% specificity, 100% sensitivity, and 98.40% f-score.
• An area under the curve (AUC) of 99% demonstrates excellent discriminative ability.
• The model's robustness and efficiency were validated against established architectures (VGG19, GoogleNet, ResNet50) and state-of-the-art methods.

Conclusions:

• The developed hybrid deep learning model effectively classifies skin lesions and accurately predicts melanoma.
• The proposed approach successfully overcomes class imbalance issues, enhancing diagnostic reliability.
• The model demonstrates significant potential for clinical application in dermoscopy and early melanoma detection.