Machine learning and deep learning are increasingly used in time series forecasting in critical fields like healthcare, finance, and agriculture, making the need for explainable artificial intelligence essential to ensure transparency and trust. In this work, a novel methodology is proposed to assess the influence of input features on the quality of explanations that describe the behavior of the models in the context of time series forecasting. To investigate this aspect, controlled perturbations are introduced to individual input features, and the resulting variations in the generated explanations are analyzed to assess their stability and reliability. The proposed methodology relies on the generation of association rules, which are inherently interpretable and capable of summarizing data patterns transparently. The results show that the proposed methodology can be applied to real-world time series data, particularly in the field of agriculture, where planning decisions frequently rely on forecasts across different time frames.