Novel Framework for Weather Prediction using Genetic Algorithm-based Model Tuning

Machine learning model performances are very sensitive to hyperparameters so it is very crucial to have hyperparameter tuning. One of the focuses of the work presented in this paper is to assess the effect of hyperparameter optimization techniques, such as grid search, random search, and genetic algorithms (GA), on weather prediction models. So, for this, this study implements five machine learning models namely linear regression, random forest, artificial neural network (ANN), feedforward neural network (FNN) and gradient boosting using a dataset of a global weather repository. The study finds that in ANN, the GA performs best in terms of finding optimal hyperparameter values to predict the time series with the minimum Mean Squared Error (MSE) and maximum R-Squared (R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>) values. The impact of different hyperparameters of the model, namely learning rate, number of neuron layers, and number of neurons on the model performance is also studied. The study also discusses the benefit of GA for exploring the huge and intricate hyperparameter space and achieving conspicuous performance enhancements over conventional ways. These results highlight the role of GAs in increasing prediction accuracy which can be used in other tasks of machine learning.