Home // DATA ANALYTICS 2015, The Fourth International Conference on Data Analytics // View article
Authors:
Phil Romero
Kalpak Dighe
Keywords: Digital Signbal Processing ; Machine Learning ; Radio Frequency
Abstract:
Collection of Radio Frequency data can overwhelm even the largest data storage capacities very quickly due to high sampling frequencies. There are many sources of possible error in maintaining an accurate record of the captured signals. These issues can be solved, in large part, through an automatic classification of data sets gathered that eliminates the possibility of human error and assures that the proper type of signals were captured in a timely fashion. In this paper, we will describe the process used to produce a classification system. The goal is to identify and use measures produced from the raw signal information and/or the spectrograms for input into an algorithm that produces clusters based on similarity that will classify the data into subsets with the least amount of computational complexity. K-means clustering and principal component analysis are utilized in a two step process to perform the classification of the data sets. Minimal amounts of measures have been found to produce satisfactory results in separating the raw signal data into dissimilar signal types based on a 32768 sample size. This minimizes computational complexity while still producing output used in the second stage of the process to classify the data sets. A method of classification was found that produces minimal false positive errors while selecting the proper number of clusters without resorting to more computationally complex methods thereby decreasing the time spent classifying.
Pages: 146 to 152
Copyright: Copyright (c) The Government of USA, 2015. Used by permission to IARIA.
Publication date: July 19, 2015
Published in: conference
ISSN: 2308-4464
ISBN: 978-1-61208-423-7
Location: Nice, France
Dates: from July 19, 2015 to July 24, 2015