In this chapter, we are going to develop a Deep Neural Network (DNN) using the standard feedforward network architecture. We will add components and changes to the application while we progress through the recipes. Make sure to revisit Chapter 1, Introduction to Deep Learning in Java, and Chapter 2, Data Extraction, Transformation, and Loading, if you have not already done so. This is to ensure better understanding of the recipes in this chapter.

We will take an example of a customer retention prediction for the demonstration of the standard feedforward network. This is a crucial real-world problem that every business wants to solve. Businesses would like to invest more in happy customers, who tend to stay customers for longer periods of time. At the same time, predictions of losing customers will make businesses focus more on decisions that encourage customers not to take their business elsewhere.

Remember that a feedforward network doesn't really give you any hints about the actual features that decide the outcome. It just predicts whether a customer continues to patronize the organization or not. The actual feature signals are hidden, and it is left to the neural network to decide. If you want to record the actual feature signals that control the prediction outcome, then you could use an autoencoder for the task. Let's examine how to construct a feedforward network for our aforementioned use case.

In this chapter, we will cover the following recipes:

• Extracting data from CSV input
• Removing anomalies from the data
• Applying transformations to the data
• Designing input layers for the neural network model
• Designing hidden layers for the neural network model
• Designing output layers for the neural network model
• Training and evaluating the neural network model for CSV data
• Deploying the neural network model and using it as an API