Enhanced Skin Cancer Detection Through Deep Learning and Random Search Optimization

Skin cancer diagnosis is a critical area of medical research due to its prevalence and potentially life-threatening nature. Conventional diagnostic approaches often depend on subjective evaluation through traditional means, which can be laborious, expensive, and time-consuming. The objective of this research endeavor is to present an innovative method for the automated classification of skin cancer by utilizing deep learning methods, specifically the EfficientNetB0 model. Initially, the default model produced a 66% testing accuracy rate on the dataset. After balancing the data and fine-tuning the EfficientNetB0 model by using the Random Search optimization method the final training accuracy of 98.98% and a test accuracy of 96.73% were attained. In addition, the model's dependable performance on unknown data along with low training and testing losses indicated that there was minimal overfitting. The model performed exceptionally well across all classes of skin cancer lesions, with a weighted average performance of 97%, according to a thorough examination that used precision, recall, and F1-score criteria. The results of this study highlight the precision and efficacy of the model in categorizing various forms of skin cancer lesions, thereby demonstrating its potential application in clinical environments to deliver precise diagnoses and devise effective treatments.