As seen in Chapters 1, Neural Network and Artificial Intelligence Concepts, and Chapter 2, Learning Process in Neural Networks, training a neural network model forms the basis for building a neural network.

Feed-forward and backpropagation are the techniques used to determine the weights and biases of the model. The weights can never be zero but the biases can be zero. To start with, the weights are initialized a random number, and by gradient descent, the errors are minimized; we get a set of best possible weights and biases for the model.

Once the model is trained using any of the R functions, we can pass on the independent variables to predict the target or unknown variable. In this chapter, we will use a publicly available dataset to train, test, and visualize a neural network model. The following items will be covered:

• Training, testing, and evaluating a dataset using NN model
• Visualizing the NN model
• Early stopping
• Avoiding overfitting
• Generalization of NN
• Scaling of NN parameters
• Ensemble models

By the end of the chapter, we will understand how to train, test, and evaluate a dataset using NN model. We will learn how to visualize the NN model in R environment. We will cover the concepts like early stopping, avoiding overfitting, generalization of NN, and scaling of NN parameters.