Abstract. Current techniques for calculating and generating models used for analyzing the Earth’s magnetic field are laborious and time-consuming. We assert that machine learning can have a significant impact on building magnetic field models more quickly and on various levels of complexity, specifically as it pertains to data cleansing and sorting. Our approach to this problem uses a reverse iterative multi-phase process for data cleansing, in which, initially, the CHAOS-6 model data is examined to determine if machine learning can be used to differentiate between useful data components for spherical harmonics, versus data noise. During this phase, six different machine learning techniques are used and compared: two classification techniques (Convolutional Neural Network (CNN) and Support Vector Classification (SVC)) and four regression techniques (Random Forest Regression (RFR), Support Vector Regression (SVR), Logistic Regression, and Linear Regression). During this initial phase, the focus is on understanding the accuracy of machine learning for model selection and uses relatively clean data. Future phases should include machine learning relevance as it pertains to the massive volume of data received from satellites. Exploring the machine learning capabilities for magnetic field datasets accomplishes 1) faster and more efficient computation when there are millions of rows of data in any given 30-day period, and 2) lowers the propagation of errors that cause some data to be useless in the spherical harmonics computations used in the model generation.
Loftin, Sheri; Fite, Sarah J.; Bishop, Laura V.; and Kotsiaros, Stavros
"Machine Learning vs Conventional Analysis Techniques for the Earth’s Magnetic Field Study,"
SMU Data Science Review: Vol. 2:
1, Article 7.
Available at: https://scholar.smu.edu/datasciencereview/vol2/iss1/7
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License