<p>Positioning systems are essential for various products and service operations. Nonetheless, due primarily to channel distortions and signal attenuation, the global navigation satellite systems do not perform satisfactorily in indoor environments. In this context, visible light positioning methods based on received signal strength show promise as candidates to meet this demand due to their straightforward implementation. However, many of these methods rely heavily on complex mathematical system modeling, which can result in poor performance under nonideal circumstances. Hence, machine learning techniques have the potential to yield superior outcomes due to their inherent ability to adapt to various scenarios. Therefore, this work employs both numerical simulation and practical implementation approaches to conduct a comparative study between the supervised learning approaches of feedforward neural networks and K-nearest neighbors and the traditional method based on the maximum likelihood estimator for the task of three-dimensional spatial localization in indoor environments, carrying out a meticulous hyperparameter tuning procedure for each estimator. The findings indicate that the machine learning approach can achieve accuracy, precision, and convergence probabilities comparable to those of model-based methods, while drastically reducing prediction times. It’s worth noting that the maximum likelihood estimator surpasses the other techniques in low-noise situations, delivering the highest accuracy and precision, with the feedforward neural network also performing well. On the other hand, K-nearest neighbors exhibit the least favorable performance among the evaluated approaches. Moreover, the proposed machine learning methods demonstrate superior performance in scenarios with elevated noise levels and real-world applications.</p>