Machine Learning for Developers

Types of problem to be tackled

Neural networks can be used for both regression problems and classification problems. The common architectural difference resides in the output layer: in order to be able to bring a real number-based result, no standardization function, such as sigmoid, should be applied. In this manner, we won't be changing the outcome of the variable to one of the many possible class values, getting a continuum of possible outcomes. Let's take a look at the following types of problems to be tackled:

• Regression/function approximation problems: This type of problem uses a min squared error function, a linear output activation, and sigmoidal hidden activations. This will give us a real value for output.
• Classification problems ...