Skin-lesion studies have achieved promising classification accuracy; however, their implementation in mobile–cloud environments remains limited. This study presents Skin Alert, a cloud-based prototype for classifying seven skin-lesion categories in the HAM10000 dataset using an ImageNet-pretrained InceptionV3 model. Hyperparameters, including learning rate, dropout rate, batch size, and number of epochs, were optimized using grid search. The original class distribution was retained, while data augmentation and class-weighted learning were applied during training. The final model achieved an internal validation accuracy of 85.32%, weighted precision of 85.79%, weighted recall of 85.32%, weighted F1-score of 85.38%, macro F1-score of 72.93%, and macro-average one-vs-rest AUC of 96.84%. The model was deployed through a Flask API on Google Cloud Platform and integrated with a Flutter application, Firebase Authentication, Realtime Database, and Storage. Black-box and backend integration testing confirmed that predefined workflows operated as intended. The locally hosted API achieved approximately 250–300 ms response time per image, while cloud deployment required approximately 1.10 s per image. These findings demonstrate the technical feasibility of the proposed mobile–cloud prototype.
Copyrights © 2026