Machine Learning Based System for Nephrologist to Classify Diabetic Kidney Disease

Progressive loss of renal function is the hallmark of diabetic kidney disease (DKD), a disorder that advances over time and carries a high death rate. Diabetic kidney disease (DKD) can be identified early and treated if found early. Since DKD doesn't have any symptoms, it raises the risk factor when it's at a critical state. In order to prevent DKD, it is crucial from a medical standpoint to use a machine-first model to anticipate when the disease may manifest. Based on their health information, people can use open-source machine learning algorithms to assess their risk of acquiring DKD. In this proposed work, the patient health care dataset is subjected to a variety of machine learning techniques. The quantity of datasets and clinical factors utilized to train the algorithm determines the range of prediction accuracy levels. The goal of this study is to increase the overall accuracy through the use of a combination of machine learning algorithms. The suggested model uses the K-Nearest Neighbour, SVM, and Hybrid algorithm -GBRF -Random Forest mixed with Gradient Boosting that allows for hyper parameter fine tuning. This project's combination strategy produced more reliable results when compared to models utilized in other investigations. Clinical data from a repository serves as the foundation for the existing model. To validate the results, real-time data was also acquired. Random Forest demonstrates the most effective performance of GBRF with the highest accuracy score of 97.5% for repository data and 100% for clinical real-time data, respectively, based on the accuracy, recall, and f1-score from the classification reports among the three algorithms combined with a gradient booster.