<p>Indoor localization using radio frequency signals presents significant challenges in manufacturing environments due to complex signal propagation conditions. This study investigates the performance of Time of Arrival (ToA) measurements for indoor localization by exploring various Access Point (AP) configurations, power levels, and bandwidths. We evaluate traditional methods, such as K-Nearest Neighbors (KNN) and Trilateration, alongside machine learning (ML) techniques, including Deep Neural Networks (DNN), Long Short-Term Memory (LSTM), and Convolutional Neural Networks (CNN), to enhance localization accuracy. Through simulations, the models are tested under controlled variations of transmission power and signal bandwidth, serving as proxies for signal quality, and different AP availability scenarios. Results demonstrate that while traditional methods perform well in ideal conditions, neural network-based models consistently outperform them, particularly under low transmission power, limited bandwidth, or sparse AP coverage. Specifically, in the lowest transmission power scenarios, machine learning models reduced the localization error by approximately 30% compared to trilateration. Among the evaluated neural networks, CNN and LSTM demonstrated superior robustness in specific high-bandwidth scenarios; however, the DNN model provided the most favorable trade-off between accuracy and computational complexity. Our results demonstrate that integrating ToA with machine learning yields robust localization in challenging environments, optimizing both operational efficiency and safety in manufacturing settings.</p>