MULTI-CLASS NUTRIENT DEFICIENCY DETECTION FROM SKIN IMAGES USING CLASSICAL MACHINE LEARNING AND DEEP LEARNING MODELS

Abstract

The purpose of deficiency detection is to provide a methodology that accurately detects deficiency in human body without using any laboratory reports. In the field of computer vision, different deep learning models have been used to detect deficiency in humans. These models can detect only one or two types of deficiencies. There is no method available that detects multiple types of deficiencies in human body. However, it is important to keep in mind that these methods have some limitations when it comes to detect different categories of deficiencies. In previous research, low accuracy is observed due to the use of insufficient data. However, in this study, we address this limitation by introducing an additional category of malnutrition. To enhance the dataset, we collect images of patients with malnutrition specifically from DHQ hospital. This approach aims to improve the accuracy of our research findings and contribute valuable insights to the field of deficiency detection. We exploit six machine learning models such as Decision Tree, Naive Bayes, SVM, ResNet50, CNN, and Vision Transformer, as we train and test them for image classification tasks. The Vision Transformer out-performs other models, achieving an exceptional accuracy of 98%. It stands out in finding detailed patterns in image data, showcasing impressive accuracy. The ongoing comparison of these models not only confirms the effectiveness of the Vision Transformer but also provides insights for utilizing transformer-based architectures in various computer vision applications