• A common practice is to split a dataset into three parts using the following percentages: 60 percent for training (773 observations), 20 percent for cross-validation (257 observations), and 20 percent for testing (257 observations).
• To optimize the predictive power of the neural network model, both the cross-validation and the training dataset were used simultaneously.
• One the predictive power of the training model reached the optimal level, the neural network weights were saved for the testing data.