3 Deep learning

Deep learning is a subset of machine learning that uses neural networks with multiple layers (“deep” architectures). Compared with many classical ML models, deep learning can capture complex nonlinear relationships and can be especially powerful for unstructured data such as images and text. In practice, the deep learning workflow still follows the same discipline as standard ML:

1. define the prediction target and the feature set,
   
2. split data into training and test sets,
   
3. preprocess features (scaling, encoding, reshaping),
   
4. build and compile a model,
   
5. train the model and monitor overfitting,
   
6. evaluate on the test set, and
   
7. review predictions and performance metrics.

In this chapter, we demonstrate deep neural networks for: - binary classification (Pima Indians Diabetes), - regression (Boston housing), - image classification using a convolutional neural network (CNN) (MNIST).

We use the keras package in R, which provides a user-friendly interface to define and train neural networks.

For more details, please read here.