A Survey of Hyperparameter Selection Methods for Weather Forecasting Using State-of-the-Art Machine Learning Algorithms

Abstract

Weather forecasting is an important aspect across various sectors, but the intricate dynamics of weather systems pose a challenge for conventional statistical models to forecast accurately. Besides auto-regressive time forecasting models like ARIMA, deep learning architectures like ANNs, LSTMs, and GRU networks have been shown to enhance the accuracy of forecasts by considering temporal dependencies. This paper studies various machine learning models like XGBoost, SVR, KNN Regressor, Random Forest Regressor and the application of metaheuristic algorithms, like Genetic Algorithm (GA), Differential Evolution (DE), and Particle Swarm Optimization (PSO), on some deep learning model architectures like ANNs, LSTMs and GRUs, to automate the process of finding the best hyperparameters for the models. Furthermore, this paper explores the Quantum LSTM (QLSTM) network and novel QLSTM Ensemble models. We conduct a comparative study of these model structures, evaluating their effectiveness in weather prediction using measures such as Mean Squared Error (MSE) and Mean Absolute Percentage Error (MAPE). The findings underscore the capabilities of metaheuristic algorithms and innovative quantum methods in enhancing the precision of weather forecasts. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.