Automated Detection of Diabetic Foot Ulcers Using Inception V3:A Deep Learning Approach for Early Intervention

This study explores a machine learning approach for classifying skin conditions into “Abnormal (Ulcer)” and “Normal (Healthy Skin)” using the Inception V3convolutional neural network. The dataset used here was sourced from Kaggle and includes high-resolution skin images of both ulcerated and healthy skin, preprocessed by resizing, normalization, and augmentation to improve model generalization. Stratification is done into training, validation, and testing subsets, ensuring the classes are balanced. In this model, transfer learning is used and fine-tuned with dropout and regularization in order to avoid overfitting. The overall accuracy is 79.72 %. Training and validation loss indicate that overfitting occurred, since validation loss is no longer improving after its first couple of reductions. Validation accuracy stands at a constantly lower rate, 76.9%, compared with training accuracy reaching up to 79.7%. The same conclusions are brought by the confusion matrix: there are definitely good performances concerning ulcers-82 % recall-and healthy skin classes-81 % precision-but improvements in FN and FP must be envisaged to yield even better results. A balanced performance, however, on all classification metrics assures reliability yet still generalizable. Future works will include, but are not limited to, regularization techniques, tuning of hyperparameters, and inclusion of clinical features to make more informed decisions.