An efficient Analysis based on the Internet of Things, SVM and KNN for Operative Diabetic Retinopathy Classification
Vishakha Vishakha,,,,,,,Arun Khatri,Abhiraj Malhotra,Mahesh TR,Jagmeet Sohal,Raenu Kolandaisamy
DOI: https://doi.org/10.54216/jisiot.130109
2024-01-01
Journal of Intelligent Systems and Internet of Things
Abstract:These days, diabetes is an incurable disease, with millions of people suffering from it worldwide. Several variables namely lack of education, crowded living conditions, obesity and improper diet are among the causes of this recent upsurge in diabetes cases. They are identified by the name of infections induced by bacteria or viruses, harmful compounds in food, autoimmune reactions, obesity, unhealthy lifestyles, and pollution in the environment. Excessive and sight-threatening diabetic retinopathy (DR) is the most common retinal micro-vascular dysfunction that is characterized by the occurrence of a disorder of retinal blood vessels resulting in impaired vision. The IoT-based work is conducted in this work on the machine learning (ML) techniques, K-Nearest Neighbor (KNN) and Support Vector Machine (SVM). The classification of diabetic retinopathy is a topic that is under research. The range of activities of the processes of downsampling, labelling, image flattening, and format conversion is all within the dataset preparation process. An advanced prognosis model is designed which follows a combination of two machine learning techniques such as SVM and KNN. This approach classifies the images of diabetic retinopathy into five segments (subclasses), thus facilitating in-depth analysis. Our solution proposal in this case is a superior one because of its higher classification accuracy and faster processing speed as the findings showed. The robustness and accuracy that the SVM is known for are ensured by the convergence of the KNN to the SVM. The paper also proves a close linkage of clinical symptoms and blood sugar readings to an algorithmic DM prediction system that is based on IoT and ML approaches. This is another advantage of this method that it outperforms the existing classification methods. Amongst all the classifiers that we used in this project, the KNN ML classifier turned out to be the most accurate one with an accuracy rate of 93%. It was found that the algorithm performed with a 79% accuracy rate after tough testing and training and it was consistently providing number one quality DM predictions.