Machine Learning with the backpropagation algorithm

Abstract

The goal of this project was to develop with Python a classiffer capable to separate two different classes (binary classiffer). To do this, we implemented a neural network that uses gradient descent with the backpropagation algorithm to learn in a supervised training process. We first trained the net and studied its behavior with a binary classification problem that we created, generating our own training examples with the acceptance-rejection algorithm. We realized that the net was returning the conditional probability for each example to belong to one of the classes. After that, we focused on two different real classification problems, both obtained from available online machine learning data sets.