Intelligent Context Fusion for Early Anomaly Detection in IoT-Driven Healthcare Systems

Abstract

The rapid growth of smart healthcare systems and interconnected environments has increased the need for secure, scalable, and intelligent data analysis mechanisms. With the rise of distributed architectures and cloud-based services, achieving accurate prediction and efficient anomaly detection remains a significant challenge. This research presents a cloud-driven intelligent system that integrates machine learning (ML) techniques with a client–server architecture to enable secure and real-time prediction. The primary issue addressed is the difficulty of achieving high prediction accuracy from large and imbalanced datasets in distributed environments. Traditional systems often depend on standalone models and manual analysis, leading to limited accuracy, poor scalability, and ineffective handling of class imbalance, making them unsuitable for real-time, large-scale applications. To overcome these limitations, the proposed system introduces an automated, robust, and scalable framework capable of efficient data preprocessing, imbalance handling, and support for multiple ML models while ensuring secure remote access and seamless communication. The architecture consists of Laptop 1 (LP1) as the server and Laptop 2 (LP2) as the client. LP1 integrates a desktop graphical user interface with a cloudbased prediction engine using the Flask framework. The system applies preprocessing techniques such as label encoding (LE) and Synthetic Minority Oversampling Technique (SMOTE) for class balancing. It implements multiple models including RidgeClassifier (RC), QuadraticDiscriminantAnalysis (QDA), and Perceptron (PC). Additionally, a Locally Deep Ensemble Classifier (LDEC) combines HistGradientBoostingClassifier (HGB) and LGBMClassifier (LGBM) using soft voting (SV). Secure authentication is achieved using Redis, improving accuracy, scalability, reliability, and security for modern applications.