Latency-Aware Encrypted Data Processing Framework for Distributed Medical Diagnostics at the Network Edge

Abstract

The rapid growth of digital healthcare systems has increased the demand for intelligent and secure diagnostic solutions capable of handling sensitive medical data. Traditionally, disease prediction relied on manual analysis and rule-based approaches, which were limited in processing large-scale and complex datasets. With the evolution of machine learning, automated diagnostic systems have emerged; however, many existing methods lack proper data security and struggle with imbalanced datasets, resulting in reduced accuracy and reliability. These challenges highlight the need for a robust and secure analytical framework. The primary objective of this study is to develop a secure and accurate disease prediction system for healthcare datasets such as heart disease and hypothyroid conditions. Traditional systems often fail due to inadequate preprocessing, poor feature handling, and absence of secure data processing mechanisms. Additionally, class imbalance further impacts model performance and prediction consistency. To address these issues, a comprehensive framework is proposed that integrates data preprocessing, exploratory data analysis, and synthetic data balancing using Synthetic Minority Over-sampling Technique (SMOTE). Multiple machine learning models, including Naive Bayes (NB), Passive Aggressive (PA), Ridge Classifier (RC), and Extreme Gradient Boosting (XGB), are utilized for prediction. A privacy-preserving encryption mechanism is incorporated to ensure secure data handling, and a Flask-based interface enables efficient user interaction and real-time prediction. The proposed system achieves high performance, with XGB attaining 100.00% accuracy on the heart dataset and 99.89% on the hypothyroid dataset. The framework ensures accuracy, security, and scalability for reliable healthcare decision support.