A Privacy-Centric Edge Intelligence Paradigm for Instantaneous Clinical Insight Generation via Secure Data Transformation

Abstract

The rapid integration of digital technologies in healthcare has intensified concerns surrounding data security and patient privacy, especially in predictive diagnostic systems. Conventional machine learning approaches typically rely on direct access to raw medical data, making them vulnerable to data breaches, unauthorized usage, and privacy compromises. Additionally, many existing solutions lack comprehensive authentication protocols and fail to safeguard sensitive information during both training and inference stages. These limitations highlight the necessity for intelligent frameworks that can deliver high predictive performance while maintaining strict data confidentiality and secure system access. To overcome these issues, this study introduces a secure and privacy-aware healthcare prediction framework that combines machine learning with encryption and authentication mechanisms. The system is implemented using the Flask web framework and integrates an email-based One-Time Password (OTP) verification process via Simple Mail Transfer Protocol (SMTP) to ensure secure user authentication. To protect sensitive medical data, a Lightweight Privacy-Preserving Machine Learning Encryption (LPME) approach is employed, which transforms input features using polynomial-based cryptographic techniques before they are utilized for model training. The framework is evaluated on healthcare datasets related to heart disease and hypothyroidism. Prior to model development, the data undergoes preprocessing steps such as normalization, categorical encoding, and class imbalance handling using the Synthetic Minority Oversampling Technique (SMOTE). For predictive modelling, Extreme Gradient Boosting (XGB) is adopted as the primary classifier due to its effectiveness in modelling complex relationships, while Gaussian Naive Bayes (GNB) is used as a baseline for comparison. The system’s performance is assessed using standard evaluation metrics including accuracy, precision, recall, and F1-score.