Comparative Analysis of Multiclass Classification Using AE-CNN Method Combination on Nail Diseases

In the field of public health, there are various ways to diagnose multiple diseases in the human body. Diagnosing diseases at an early stage is essential. Multiple parts of the human body are observed to diagnose diseases, including the nails. The development of technology such as machine learning has brought many benefits in the health field, especially in disease detection, including the ability to detect nail diseases using nail images. This research aims to design and build a classification model for nail diseases such as Melanonychia, Nail dystrophy, normal, Onycholysis, and Onychomycosis. This research develops an AECNN model that combines an Autoencoder (AE) with a Convolutional Neural Network (CNN) with its architecture to perform multiclass classification of nail diseases. Each modeling in this study trains the model using images with noise to train the modeling on poor-quality photos. Then, the performance of the AE-CNN model was compared with several other models, namely AE-VGG16 and AE-ResNet50, based on accuracy, precision, recall, and F1-Score values. This study found that the proposed AE-CNN model was able to classify nail diseases better than the other models, with an accuracy of 91.8% for AE-CNN, 83.5% for AE-VGG16, and 48.3% for AE-ResNet50.