ANALISIS PERFORMA MODEL CNN MELALUI EVALUASI EKSTERNAL PADA SISTEM KLASIFIKASI CITRA MAMMOGRAFI BERBASIS WEBSITE LOKAL

The performance of a Convolutional Neural Network (CNN) model for mammographic image classification needs to be evaluated using data that were not involved in model development to assess its ability to generalize to data from different sources. This study aims to analyze the performance of a CNN model through external evaluation on a local web-based mammographic image classification system. The method involved applying the previously developed EfficientNetB2 model to 15 clinical mammographic images as external data. Each image was entered into the system to obtain its classification result, which was then compared with the diagnostic label in the original data. During the model development stage, the EfficientNetB2 model achieved an accuracy of 72%, precision of 67%, recall of 96% and an F1-score of 79%. The external evaluation results showed that 13 out of 15 images produced predictions consistent with their diagnostic labels, while 2 images produced inconsistent predictions, resulting in a prediction agreement rate of 86.67%. Based on the confusion matrix, the model achieved an accuracy of 86.67%, precision of 80%, recall of 100%, and an F1-score of 88.89%. The external evaluation provides an overview of the model’s generalization capability when applied to data outside its development environment through a local web-based system.