<![CDATA[Breast cancer is one of the leading causes of death among women, especially in developing countries like Indonesia. Early detection is very important to increase the cure rate and decrease the mortality rate. This research aims to improve its ability to identify cancer tumors by maximizing the recall value and comparing various transfer learning-based Convolutional Neural Network (CNN) models to find the most optimal model. The CNN architectures studied in this research include MobileNetV2, ResNet50, VGG16, and AlexNet. All models were applied to the BreastMNIST dataset, which consists of ultrasound images with two classes, benign and malignant. Transfer learning was used to overcome the challenge of limited availability of pre-labeled medical image data. The model performance was thoroughly evaluated using accuracy, precision, recall, F1-score and Area Under Curve (AUC) metrics. The results showed that MobileNetV2 provided superior performance with an accuracy of 91.14%, recall of 94%, precision of 93%, F1-score of 94%, and AUC of 0.9607. These findings indicate that the transfer learning-based MobileNetV2 is highly effective in detecting malignant tumors and is the most optimal architecture in this study.]]>
Copyrights © 2025
