Document Type

Thesis or dissertation

Date of this Version

2018

Advisor

Paul Shaman

Abstract

Historically, traditional methods such as Autoregressive Integrated Moving Average (ARIMA) have played an important role for researchers studying time series data. Recently, as advances in computer science and machine learning have gained widespread attention, researchers of time series analysis have brought new techniques to the table. In this paper, we examine the performance difference between ARIMA and a relatively recent development in the machine learning community called Long-Short Term Memory Networks (LSTM). Whereas many traditional methods assume the existence of an underlying stochastic model, these algorithmic approaches make no claims about the generation process. Our primary measure of performance is how well each model forecasts out-of-sample data. We find that data with strong seasonal structure are forecast comparatively well by either method. On the other hand, without strong seasonality, there is very little information that can be extracted and both methods tend to perform poorly in forecasting.

Keywords

time series, ARIMA, LSTM, forecasting

Share

COinS
 

Date Posted: 26 June 2018

 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.