A Validation Study of Time Series Data Forecasting Using Neural Networks

Abstract

Artificial Intelligence (AI) is a growing topic in Computer Science, and has many uses in real world applications. One application is using AI, or more specifically Neural Networks, to model data and predict outcomes. Neural Networks have been used in the past to predict weather changes, create facial recognition software, and to create self-driving cars. Our project is a validation study of, “Modeling Time Series Data With Deep Fourier Neural Networks” by Gashler and Ashmore, 2016. Gashler and Ashmore trained a Deep Fourier Neural Network to fit time-series data, such as weather. Their method was demonstrated with the weather data obtained from Anchorage, Alaska over a five year period beginning in April of 2009. In our research, we attempt to fit a simpler neural network to the Alaska weather data. We first fit our data with traditional neural network training algorithms. We then expand our research to use machine learning to train the neural network to the Alaska weather data. We believe that we can create a simpler neural network that is still as effective as Gashler and Ashmore’s neural network. This validation study provides a useful introduction to Artificial Intelligence for upper level undergraduate students. Completed code will be made public on the SWOSU Github repository. As computing resources and programming environments continue to improve, the value of forecasting will continue to increase. One may see this research as a way to improve famili