A Secure Blockchain-based Patient-centric Electronic Health Record System for Diabetes Detection Using Machine Learning

Abstract

Diabetes is now an important global health problem. This paper presents a patient-oriented electronic health record system for predicting diabetes with machine learning. The diagnosis given by the doctor and the patient ID will be written into the Ethereum Smart Contract. Algorithms analyzed and compared using the Pima Indians Diabetes Dataset are Support Vector Machine, Decision Tree, Random Forest, Extreme Gradient Boosting, and LightGBM. We have also used a decentralized, tamper-proof storage, InterPlanetary File System (IPFS), for storing the patient reports off-chain, and we will store the corresponding hash generated by the SHA-256 cryptographic hash algorithm in the smart contract, having gas optimization as well as scalability. Patients will use the Proof of Stake consensus mechanism-confirmed transparent smart contract transactions to grant or withdraw authorized user access permissions to their health data, with a One-Time Password (OTP) allowing for additional security. In our experimental result, the Support Vector Machine yielded the highest accuracy of 85.06%, f1-score of 0.7928, and a recall of 0.8148, outperforming all other algorithms. This work significantly contributes to privacy preservation by enhancing the existing approaches for diabetes diagnosis in healthcare while keeping the privacy of patient data.