The contextualised application of Explainable Artificial Intelligence (XAI) methods is a critical factor in addressing regression-based problems. However, existing XAI approaches are often insufficiently developed for such specialised usage. This study investigates and optimises the functionality of a novel technique-the Mean Background Method-which is designed to enable the contextualised application of XAI methods for time-series machine learning models and regression-based problems. The proposed method was tested on a predictive model for temperature estimation, leveraging a large multivariate time-series dataset. The method was meticulously parameterised to maximise the stability of feature importance results, ensuring robust and reliable interpretations. Experimental evaluations demonstrate that the Mean Background Method can be effectively used for clear and stable explaining of model behaviour. Furthermore, the study highlights the method’s potential to generalise to other regression-based, context-specific problems, emphasising its utility in real-world applications. This work represents a significant step forward in bridging the gap between XAI techniques and their practical application to domain-specific challenges, offering a reliable framework for enhancing model transparency and decision-making in time-series analysis.