Performance analysis of classification models to determine the health status of edge computing devices

Abstract

Artificial intelligence (AI) has contributed to the development of autonomous systems in the healthcare field by integrating machine learning models, whose evaluation on resource-limited hardware devices is important to ensure their efficiency. This research evaluates the performance of classification models in edge computing (EC) systems, considering metrics such as accuracy, latency, memory consumption, and energy efficiency on low-power microcontrollers using TinyML techniques. The processes involved include the development, implementation, and testing of algorithms on embedded hardware using differentiated preprocessing techniques and the validation of hypotheses through statistical analysis. The results show that the decision tree (DT) model is more efficient in terms of prediction time and energy consumption, while random forests (RFs) stand out for their greater accuracy. Furthermore, memory analysis reveals that models based on fully connected neural networks are more efficient in terms of RAM usage. This provides guidelines for selecting algorithms in resource-constrained environments.