This work explores the use of machine learning techniques for estimating critical Adaptive Optics (AO) parameters—Strehl ratio, seeing, wind speed, and outer scale—from Wavefront Sensor (WFS) telemetry data. These parameters are essential for assessing AO performance and understanding atmospheric turbulence. Analytical approaches exist, but often face limitations in accuracy, particularly for parameters like the outer scale. To overcome these challenges, we employ Recurrent Neural Networks (RNNs) trained on simulated telemetry data from the SOUL AO system at the Large Binocular Telescope (LBT). The results show that our machine learning approach outperforms traditional methods, and future work will aim to apply this methodology to real-world data.