Advancing Hair Disease Diagnostics: A Deep Learning Approach Using Inception-ResNet v2 for Multi-Class Classification

This research study focuses on the development of a high-performance deep learning model in relation to the precision classification of hair diseases as a means of meeting the rising requirements for automated dermatology diagnostic tools. Using the Inception-ResNet v2 architecture, this work shall try classifying 10 different hair disease classes: from Alopecia Areata, Psoriasis, to Male Pattern Baldness, with a balanced dataset of 12,000 images collected from Kaggle. Extensive preprocessing was done on the dataset, which included resizing, normalizing, and augmentation of data to improve richness and generalization of the model. The data was split into training sets and validation and testing subsets to make sure that the model is adequately evaluated as it contains 9600, 1200, and 1200 images each. This Inception-ResNet v2 has been fine-tuned by transfer learning from the pre-trained ImageNet network. Other layers added were Global Average Pooling, dense layers, batch normalization, and dropout for robustness. It was trained for 20 epochs with callbacks given for EarlyStopping and ModelCheckpoint to avoid overfitting and also to optimize performance. The general model-achieved accuracy in the proposed model is 94.7%. The proposed model has balanced precision, recall, and F1-scores for each class, standing at 0.94. Strong performances are found for Folliculitis and Head Lice, while minor misclassifications occur for the overlapping conditions, including Male Pattern Baldness and Seborrheic Dermatitis. This research shows the capability of deep learning for dermatological diagnostics, providing reliability to an automated solution. Further work should be pursued in expanding the data sets and interpretability of the models to eventually be able to implement these approaches clinically.