Optimization of Skin Cancer Detection to Improve Accuracy with the Application of Efficient Convolutional Neural Network and EfficientNetB2 Models

Abstract

Skin cancer, related to ultraviolet light exposure, can be caused by genetic mutations in skin cells. Risk factors include family history, fair skin, moles, weak immunity and solar keratosis. Diagnosis involves skin examination and biopsy. Previous studies using Convolutional Neural Network (CNN) successfully classified skin cancer with an accuracy of up to 99%. The process of developing an image classification model involves structured steps. The dataset is divided into training, validation, testing. Data augmentation is performed with ImageDataGenerator to enrich the dataset. CNN model (EfficientNetB2) was customized, trained for 50 epochs. Evaluation of testing data model, metrics such as loss, accuracy, precision, recall, and F1-score. Visualization of Classification Reports, Confusion Matrix. The entire process ensured in-depth analysis of the model's performance and focused attention on prediction. The study used Kaggle's secondary data "HAM10000 Preprocessed Data" with 11644 data and three attributes, showing variations before normalization. The CNN model peaked at the 8th epoch with 86% accuracy, but there is a risk of overfitting. Evaluation using Classification Report and Confusion Matrix provides details of model performance on each skin cancer class, supporting diagnosis and treatment.This article highlights the positive impact of using the EfficientNetB2 Model on skin cancer detection through an efficient Convolutional Neural Network (CNN). The model is optimal in size and structure, resulting in superior feature representation. Evaluations including Classification Report and Confusion Matrix demonstrate the model's superiority in classifying skin cancer types, particularly in the ''akiec'' and ''bcc'' classes, with high accuracy, significantly improving detection performance.