Abstract:
Chronic Kidney Disease (CKD) continues to pose a significant healthcare challenge, especially in rural areas of developing countries like Bangladesh, where access to affordable and effective diagnostic services is extremely limited. Early detection of CKD is crucial for slowing disease progression and improving patient outcomes. However, the diagnostic methods currently available are often expensive and technologically advanced, making them inaccessible to rural populations. To overcome these challenges, this thesis explores the application of machine learning (ML) models to enhance CKD diagnosis in a cost-effective manner, specifically targeting rural communities with limited healthcare resources. The primary objective of this study is to leverage machine learning to improve the accuracy of CKD diagnosis in settings with small and imbalanced datasets. Many existing ML models are trained on datasets that are predominantly composed of healthy individuals, resulting in high accuracy but poor sensitivity, leading to missed diagnoses of CKD patients. To address this, we implemented data balancing techniques and fine-tuned hyperparameters to enhance the performance of the models for accurate CKD detection. After optimizing the Support Vector Classifier (SVC), we achieved an impressive 95% accuracy with an AUC of 0.9952. Logistic Regression also performed well, reaching 97% accuracy and an AUC of 0.9986. The Random Forest classifier outperformed all other models, achieving perfect classification with 100% accuracy and an AUC of 1.0. These results suggest that optimized machine learning models hold great potential as a low-cost, accurate, and accessible strategy for early CKD detection, particularly in rural regions of Bangladesh where healthcare services are scarce. By implementing such models, it is possible to significantly improve patient outcomes while reducing the financial burden on healthcare systems.
