  Moving into the Frequency Domain with the Fourier Transform  Abstract:

Data preprocessing is an important part of data science, even more so when dealing with time series. Signal processing and the Fourier Transform can move your time series in frequency domain, where events are easier to detect. Learn how to apply the Fourier Transform to build a classification solution on top of IoT Time Series data. In this tutorial, Corey will introduce the theory behind the Fourier Transform and demonstrate how to build a classification model in the Frequency Domain by using KNIME Analytics Platform.

Session Outline:
Lesson 1:
To start this workshop we will recap Time Series data and its relation to IoT data, discuss the similarities between Time Series and Signal Analytics, and review first steps to verifying these data types are clean and ready for the Fourier Transform.

Lesson 2:
To apply the Fourier Transform we should understand intimately what it does; to do this we’ll start by reviewing its original integral form before shifting to the discrete version to be used on the digital data in the example use case - classifying audio recordings.

Lesson 3:
Application of the Fourier Transform shifts data from the Time Domain to the Frequency Domain. Doing this generates cross sectional data that can be used in classification models, but is typically of a very high dimensionality. We’ll introduce the concept of frequency bins to aggregate this data to a more practical size.

Lesson 4:
With the original signal data converted into cross sectional data and with its dimensionality reduced, we move on to discussing some common types of models that perform well on this type of data. Finally we build a Tree Ensemble model to predict the source of 2 second sections of the audio recordings.

Background Knowledge:
Familiarity with basic calculus and time series data is helpful but not required.

Bio:

Corey studied Mathematics at Michigan State University and works as a Data Scientist with KNIME where he focuses on Time Series Analysis, Forecasting, and Signal Analytics. He is the creator and instructor of the KNIME Time Series Analysis course, author of the e-book: Alteryx to KNIME, creator of the KNIME Time Series Analysis components, and Co-Author of the upcoming Codeless Time Series Analysis Book with Packt.

###### Open Data Science

Open Data Science
Cambridge, MA 02142
info@odsc.com