Disease Classification in Dermatology: A CNN-RF Hybrid Approach

The utilization of automated illness classification is of great significance in contemporary healthcare, as it has the capacity to improve the precision of diagnoses and optimize the process of clinical decision-making. This study introduces a hybrid methodology that combines the capabilities of Convolutional Neural Networks (CNN) and Random Forest (RF) to effectively classify dermatological illnesses. The dataset used in this study is extensive, consisting of 15,000 photos that are categorized into four unique classes: acne, hair loss, nail fungus, and skin allergy. The CNN component of our model, which has been fine-tuned from a pre-trained architecture, functions as a feature extractor. On the other hand, the RF algorithm is employed for ensemble-based decision-making. The hybrid model exhibits a noteworthy overall accuracy of $\mathbf{9 6 . 0 8 \%}$, showcasing its efficacy in the realm of multi-class categorization. An in-depth examination, encompassing a confusion matrix and a performance parameter table, provides insight into the model’s proficiency in accuracy, precision, recall, and F1-score for every disease category. In addition, conducting a comparison with the most advanced methods in the field highlights the competitive nature and significance of our model in the realm of dermatological disease classification. The hybrid CNN-RF model exhibits a harmonious combination of precision and comprehensibility, suggesting its capacity to bring about a transformative impact on disease detection. This advancement holds promise for healthcare professionals and those seeking medical assistance alike.