Wireless communication systems are increasingly used in industrial environments where reliable and stable connectivity is crucial. However, due to the open medium of wireless links, signal transmission is affected by interference, noise, and dynamically changing environmental conditions. These factors make the prediction of link quality a challenging task. This thesis examines how neural networks can be employed to forecast both the short-term and long-term behavior of wireless link quality by analyzing historical data. The proposed approach does not directly rely on raw signal samples. Instead, it transforms the signals into a set of Exponential Moving Averages (EMAs) computed at different smoothing scales.