EfficientNetB0-based Skin Burn Classification with Explainable AI: A High-Accuracy Approach for Medical Applications

Severe skin burns often need immediate medical attention; it is a common problem. Burn severity is considerably influenced by the extent of injury. The total body surface area (TBSA) affected is a key determinant. Burns are categorized into several special first, second, and third-degree classifications. Burn classifications are determined by severity and include a fourth-degree category. This paper introduces a highly effective novel classification model that uses EfficientNetB0. The burn degree classification was precisely and comprehensively assessed and thoroughly implemented using a curated dataset of 4538 images. The dataset comprises a large number of images sourced from Kaggle and Google Images. For an exceptionally careful quantification of the data images within the dataset, the precise count is 4682. 3575 images are classified into a training set, along with exactly 907 images used for a strict testing procedure. This model uses Explainable AI (XAI) to clarify its burn wound predictions. With 94.4% accuracy, the model proved reliable and stable enough for many real-time medical applications.