ZIP Codes and Neural Networks: Machine Learning for Handwritten Number Recognition

• Activity Number: 445
• Type: Contributed
• Date/Time: Tuesday, August 2, 2016 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Cuixian Chen* and Taylor Harbold and Courtney Rasmussen and Michelle Page
• Companies: The University of North Carolina at Wilmington and The University of North Carolina at Wilmington and The University of North Carolina at Wilmington and The University of North Carolina at Wilmington
• Keywords: Neural Networks ; Zip code data

Abstract

Neural Network is an idea from neuroscience that dates back to the 1940s. It started as using electrical circuits to model how neurons work, and has since then led to amazing advances like artificial intelligence. Neural networks utilize an oversimplification of the synapse processes that occur in the brain to interpret information. Raw input is taken in, organized and interpreted a certain way, and then a conclusion is come to. In statistics, neural networks mimic these processes by employing the methods of projection pursuit regression and back propagation. It is executed by taking linear data and putting it through complex, non-linear equations that improve themselves and get better at interpreting data with practice, just like our brains. By doing this, we create a simpler method for solving complex problems. Although neural networks have a wide array of uses, like facial recognition and stock prediction, we applied these techniques to predict the true value of hand-written digits from zip code data. Using the RSNNS package in the program R-Studio, we created prediction models that can "read" human handwriting.