Neural networks: methods and use cases for official statistics

Neural networks applied to official statistical data can have many useful applications. We propose a quick introduction to neural networks, from their theoretical foundations to their practical implementation in R and Python on specific official statistics issues. We illustrate their possibilities and limitations through three detailed use cases: 1. imputation of missing values in a survey, an important challenge for official statistics, for which predictive performance is central. 2. Image files usage, expanding the potential use of such files as statistical data. 3. Dimension reduction, which synthesises large data files and opens the way to many applications. This document is accompanied by codes to implement the methods presented.