Unleashing the power of Manta Rays Foraging Optimizer: A novel approach for hyper-parameter optimization in skin cancer classification

Abstract

Optimizing hyperparameters is crucial for improving the performance of deep learning (DL) models, especially in complex applications like skin cancer classification from dermoscopic images. This study introduces a novel hyperparameter optimization strategy using the Manta Rays Foraging Optimizer (MRFO). A model tailored for skin cancer classification is created by fine-tuning a Convolutional Neural Network (CNN) with MRFO, coupled with in-depth image preprocessing. Empirical evaluations on diverse datasets (ISIC, PH2, HAM10000) showcase the significant superiority of the MRFO-based model over conventional optimization algorithms. The model achieves impressive accuracy and loss metrics (ISIC: 99.43 %, 0.0250; PH2: 99.96 %, 0.0033; HAM10000: 97.70 %, 0.0626), outperforming alternative optimization algorithms such as the Grey Wolf Optimizer (98.33 % accuracy, 0.17 loss), Whale Optimization Algorithm (96 % accuracy), Grasshopper Optimization Algorithm (97.2 % accuracy), Densnet121-MRFO (99.26 % accuracy), InceptionV3 with GA (99.9 % accuracy), and African Vulture Optimization Algorithm (92.7 % accuracy). The novel approach demonstrates superior accuracy and loss metrics, underscoring its potential for precise and efficient skin cancer detection. Additionally, narrow confidence intervals and balanced precision-recall confirm the model’s generalizability and effectiveness, paving the way for early and accurate skin cancer detection and potentially improving patient outcomes.