Explainable Machine Learning on Classification of Healthy and Unhealthy Hair
Weng Yan Chow, Wei Wei Heng, Nurul Ashikin Abdul-Kadir, Hrikeshraj Nadaraj
Abstract
Early scalp disease diagnosis is made easier with the aid of Deep Learning (DL) algorithms. Utilizing DL algorithms eliminates the requirement for manual data reconstruction and feature extraction for classification purposes. Moreover, the growing usage of DL for essential applications such as medical diagnostics raises reliability issues and there is an urgent need to better understand the DL model predictions. In this study, we implemented a Convolutional Neural Network model to classify hair with and without scalp diseases based on online image datasets. Besides obtaining the standard performance metrics such as accuracy, precision and recall, the Local Interpretable Model-Agnostic Explanations (LIME) technique was used to interpret the model’s decisions. Our studies achieved a better accuracy of 96.63% compared to the previous studies but LIME interpretation of the results revealed that the high classification accuracy does not coincide with the model applicability or practicality.