Machine Learning in Wireless Link Estimation: A Comparative Study of Supervised, Unsupervised, and Deep Learning Approaches

The proliferation of Internet of Things devices demands reliable wireless connections against negative environmental factors and signal interference. In this paper, we present a comprehensive comparative analysis of about 17 machine learning algorithms with different settings, including unsupervised, supervised, and deep learning models, for wireless link estimation. We evaluate the accuracy and adaptability of these algorithms under various connection conditions in two datasets: the publicly available dataset from Colorado and a custom dataset collected using Raspberry Pi 4 devices. The key finding reveals that while deep learning models achieve superior performance, they are more prone to overfitting than traditional machine learning approaches. Notably, unsupervised models struggle to find meaningful cluster structures in complex datasets but achieve high accuracy on simpler ones. Based on this finding, researchers in this field can have insights into the strengths and limitations of machine learning approaches, offering a practical foundation for developing more effective and adaptable algorithms for wireless link estimation in diverse IoT environments.