Comparative Study of Regularization and Mini-Batch Training Techniques for ANN Based Wind Speed Prediction

Wind Speed (WS) prediction is a complex task due to its non-linear and dynamic nature. Machine learning models are commonly used for this purpose therefore prediction accuracy of the models remains a challenging problem now a days. In this study L2, L1 regularization, early stopping and batch normalization are used for comparative analysis in WS prediction. It is evaluated by Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R-squared ($\mathrm{R}^{2}$). For tasks with similar characteristics, prioritize using L1 Regularization as it demonstrates the best balance of error reduction and generalization. Re-evaluate the use of Batch Normalization and Early Stopping for this context, possibly in conjunction with other techniques, to improve their effectiveness.