Publication Date
2024
Document Type
Thesis
Committee Members
Fathi Amsaad, Ph.D. (Advisor); Huaining Cheng, Ph.D. (Committee Member); Wen Zhang, Ph.D. (Committee Member); Clintoria Williams, Ph.D. (Committee Member)
Degree Name
Master of Science (MS)
Abstract
Chronic Kidney Disease (CKD) poses significant health and financial threat to millions of patients all around the world. The irreversible nature of this disease not just leads to comorbid diseases like Diabetes Mellitus, Hypertension, Anemia, Bone Disease, Neurological Implants etc. It can permanently damage the kidney by progressing to Acute Kidney Injury (AKI) or End Stage Renal Diseases (ESRD). The risk factors of CKD become more dangerous as patients suffering from it have little to no idea about the presence of CKD in their body until it takes the shape of AKI or ESRD. There are severe economic burdens for the chronic kidney disease as well. Patients suffering from CKD would naturally have less productivity and hence affect the workflow. Patients admitted to hospitals for AKI or ESRD have to go through huge costs for diagnosis, treatment, hospitalizations, medications or in the final stage kidney transplantation. The dialysis and transplantation process also takes up huge costs for surgery, immuno-suppressive medications, follow-up care etc. Moreover, people struggling with financial aspects in developing countries could barely afford this large economic challenges associated with CKD. To find out which features from a patient’s demographic and clinical tests contribute in what capacity in identifying the risk factors of CKD, the scope of this research lies in application of Explainable AI (XAI) in explaining the decision-making process of ensemble models. The primary research question is what factors with what values contribute towards making CKD or Non-CKD prediction, what changes need to happen for the contributing features for a contrasting class prediction and which models would be better suited in terms of accuracy metrics and interpretability measures. The goal of this research would be to assist the clinicians as end users in explaining risk factors in CKD progression so that they can inform the patients about diet, lifestyle changes to prevent the progression to next stages. Because so many risk factors of CKD include diabetes, hypertension, obesity etc. are modifiable through diet and lifestyle changes. To address this research question, our dataset was collected on 400 subjects from University of California Irvine's Machine Learning repository. The dataset had a record of patients where 250 of them had CKD and others were healthy individuals from Apollo Hospital, India. The current scope of this research has also been compared with existing research in terms of Accuracy, Interpretability and Fidelity measures. The research findings were validated and also corrected by nephrologists for further research directions.
Page Count
80
Department or Program
Department of Computer Science and Engineering
Year Degree Awarded
2024
Copyright
Copyright 2024, all rights reserved. My ETD will be available under the "Fair Use" terms of copyright law.
ORCID ID
0009-0008-5308-4881
