Stationarity is a preprocessing step in time series analysis, it’s indispensable for traditional models, although the stationarity is not strictly required for certain advanced models. This study investigates its influence on advanced forecasting techniques, including Double Vector Quantization based on SOM (DVQ-SOM) and recurrent neural network models. These models were applied to Mauna Loa Carbon Dioxide dataset. The stationarity of the dataset is evaluated using rigorous statistical tests. This preprocessing improved the performance of the models resulting from the calculation of the Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) metrics, calculated on both stationary and non-stationary dataset. The DVQ-SOM model showed the most substantial improvement, demonstrating its strong adaptability to stationary data.