Advanced Skin Cancer Classification Using Yet Another Hill Climbing

Abstract

Abstract Hill-climbing is a heuristic search or informed search technique having different weights or variants used for mathematical optimization problems in the field of artificial intelligence. This paper proposes a new hill-climbing technique to achieve the highest prediction accuracy in the skin cancer benchmark. Changing the termination condition forces the system state to escape from the local minima and reach a better solution. The effect of the termination condition changes is similar to the hill-climbing function. The termination conditions using baseline and epochs significantly play a key role in improving prediction accuracy. This paper demonstrated the effectiveness of the proposed hill-climbing method with data augmentation. Data augmentation can convert a severe imbalanced dataset to a balanced dataset. The proposed method achieves the highest prediction accuracy, more than 0.99 prediction accuracy in classification of seven skin cancers using an imbalanced HAM10000 dataset. This paper shows the effectiveness of epoch-baseline termination management and data augmentation in machine learning. With HAM10000 dataset, the proposed advanced skin classification achieved the highest prediction accuracy in the world. The robustness of the proposed algorithm was justified by repetitive random validation and the confusion matrix. The proposed hill climbing can be used for deep learning in general.