Application of Homomorphic Encryption in Machine Learning Based Chronic Kidney Disease Prediction

Chronic Kidney Disease (CKD) is a silent but progressive illness that can lead to kidney failure if not detected early. Machine learning (ML) classifiers offer a promising avenue for early detection, but the sensitive nature of healthcare data requires stringent privacy measures. This paper proposes a novel approach that leverages homomorphic encryption to securely predict CKD using ML algorithms. We first conduct a literature review on homomorphic encryption in healthcare systems and CKD prediction using ML. We then preprocess the CKD dataset, which includes 400 samples with 24 features, and apply feature selection using mutual information. Next, we build a logistic regression model on plaintext CKD training data and use it to conduct predictive analysis on encrypted test data. We encrypt the test data using Paillier homomorphic encryption, enabling computations on the encrypted data without decryption. Our experimental results demonstrate the effectiveness of the proposed approach. Using the logistic regression model trained on plaintext data, we achieve accuracies of up to 98.75 % on encrypted test data, while preserving patient privacy. This study highlights the potential of homomorphic encryption in enhancing patient privacy and data security in healthcare applications, particularly in CKD prediction.