Designing the Output Layer

You can use a neural network both for a classification problem and for a regression problem. However, you must change the activation function of the output layer, depending on which of the problems you use a neural network for. Usually, an identity function is used for a regression problem, and a softmax function is used for a classification problem.

Note

Machine learning problems can be broadly divided into "classification problems" and "regression problems." A classification problem is a problem of identifying which class the data belongs to—for example, classifying the person in an image as a man or a woman—while a regression problem is a problem of predicting a (continuous) number from certain input data—for example, predicting the weight of the person in an image.