Given a time-series dataset and deep learning based time-series forecasting method, how can we determine the optimal hyperparameter value efficiently? How can we determine this value without running the method with a large number of candidate values? In this study, we present \textit{Xtune}, an efficient and novel hyperparameter tuning method, utilizing explainable AI, for a time-series forecasting using deep learning. Our proposed method has the following properties: (a) It is effective: it determines the optimal hyperparameter value of deep learning based time-series forecasting methods. (b) It is efficient: it does not need to run the method on various hyperparameter values. (c) It is applicable: it can be used to tune the hyperprameter tuning for the anomaly detection method utilizing deep learning-based time-series forecasting. Extensive experiments on real datasets and time-series forecasting methods demonstrate that \textit{Xtune} does indeed determine the optimal hyperparameter value efficiently, and consistently outperforms the existing methods in terms of execution speed.