Effect of Epoch and Balanced Dataset on Acne Severity Classification using MobileNetV2

Acne vulgaris is a common dermatological condition affecting individuals across various age groups, with severity assessment traditionally reliant on expert clinical evaluation that may not always be accessible. This research investigated the impact of epoch variation and dataset balancing on acne severity classification using MobileNetV2, a lightweight convolutional neural network architecture. The study utilized a balanced dataset consisting of facial images categorized into three severity classes, i.e. mild, moderate, and severe. All images underwent preprocessing, including resizing to <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$224 \times 224$</tex> pixels, to meet the input requirements of MobileNetV2. The dataset was proportionally split for training and testing phases to ensure robust evaluation. The model was trained over 460 epochs, with performance monitored through accuracy, precision, recall, and F1-score metrics. Results indicated a steady increase in training and validation accuracy, culminating in a validation accuracy of approximately 83\%. Among the severity levels, the severe class achieved the highest F1-score of 0.93, highlighting the MobileNetV2 strong capability in accurately identifying severe acne cases. The balanced dataset facilitated unbiased learning across all classes, mitigating common issues of data imbalance that often skew MobileNetV2 predictions. These findings demonstrate the effectiveness of MobileNetV2 in classifying acne severity levels and underscore the importance of balanced datasets and appropriate epoch selection in training deep learning models.