Kalman Filter (KF) is a best state estimation algorithm, which has been proven useful in various forecasting domains. One of the domains where researchers have preferred KF is weather forecasting due to its iterative blending of the previous measurements. However, its performance depends heavily on properly tuning its parameters, particularly the process and measurement noise covariances. The accuracy of prediction falls short due to the non-linearity of the weather data. The vast amount of weather data available is non-linear, complex, and uncertain. This data needs to be trained properly with appropriate parameters tuned. The main objective of this research is to find efficient ways to optimize the KF algorithm in a weather forecasting context. One of the methods for handling and understanding complex data is the hyperparameter tuning of the model. This paper focuses the need for hyperparameter tuning for the Kalman filter in weather forecasting. The models used for this research were the classical KF, and Grid search and neural network to optimize it. The results highlighted that hyperparameter tuning using the neural network method improved the model's predictions over the classical KF model. The grid search-based hyperparameter tuning showed the best results in this context. This study's results revealed the need for hyperparameter tuning and demonstrated how different tuning approaches can affect the model differently.