Comparativo de desempenho e acurácia de técnicas de Machine Learning aplicadas a ECG em infraestruturas de computação em nuvem

Abstract

The application of Artificial Intelligence (AI) techniques to support medical diagnosis has emerged as a promising approach for the automated analysis of biomedical signals, particularly in scenarios that require fast and reliable responses. In this context, cloud computing represents a viable solution to overcome the processing and storage limitations of resource-constrained devices, such as those employed in Internet of Things (IoT)-based remote health monitoring systems. This work aims to analyze the quality of cloud computing services in the execution of AI models applied to the processing of electrocardiogram (ECG) signals, considering criteria such as response time, latency, operational cost, and ease of deployment. Clinical data from the Heart Disease Dataset, provided by the UCI Machine Learning Repository and widely used as a benchmark in cardiovascular risk prediction studies, were employed in the experiments. The proposed methodology includes data preprocessing and exploratory analysis, machine learning model development and evaluation, and the implementation of an API to assess performance across different computational infrastructures. The models evaluated were KNN, Neural Networks, Support Vector Machine (SVM), and Logistic Regression. Each model was executed five times using independent data splits, and the results were aggregated using the arithmetic mean. Performance was assessed using precision, recall, F1-score, accuracy, response latency, and operational cost. The results show that the Logistic Regression model achieved the best overall performance, presenting high average accuracy and greater stability across executions. Furthermore, the comparative analysis demonstrated that cloud-based infrastructures outperform local processing in terms of latency and scalability, with AWS standing out for lower response times and Magalu Cloud for its cost-effectiveness. These findings indicate that cloud computing services are a suitable and efficient solution for AI-based diagnostic support systems, especially for time-sensitive healthcare applications.