A Multi-Model Machine Learning Framework for High-Fidelity Diagnostic Support in Emergency Medicine

Abstract

The growing occurrence of hypertension and diabetes has increased the demand for intelligent and scalable healthcare systems that support early detection and timely medical decisions. Managing these conditions requires continuous monitoring of multiple patient factors, which is challenging to perform using manual evaluation alone. Traditional systems rely on basic statistical methods and human interpretation, making them inefficient for handling large datasets, complex feature relationships, and real-time prediction requirements. As a result, such systems often lack accuracy, consistency, and the ability to operate effectively in distributed environments. There is a strong need for an automated system capable of delivering accurate predictions while enabling real-time communication between devices, particularly in emergency or distributed healthcare scenarios. To address these challenges, this research proposes a real-time Artificial Intelligence (AI) powered decision support system based on a two-laptop client–server architecture. The server system performs data preprocessing, model training, and prediction using Machine Learning (ML) models such as Complement Naive Bayes (CNB), Multinomial Naive Bayes (MNB), Perceptron, and a proposed TAO Tree Classifier. Preprocessing techniques include Label Encoding and K-Means Synthetic Minority Oversampling Technique (KMeans-SMOTE) for handling categorical data and class imbalance. A Flask-based Application Programming Interface (API) enables communication between systems using Hypertext Transfer Protocol (HTTP). The client system allows users to upload datasets, which are sent to the server for prediction of blood pressure category and diabetes status. Lightning Memory-Mapped Database (LMDB) is used for secure and efficient user data management.