Home // International Journal On Advances in Telecommunications, volume 10, numbers 1 and 2, 2017 // View article
Authors:
Taylor K. Larkin
Denise J. McManus
Keywords: ensemble modeling; space weather; quantile regression; stacked generalization; telecommunications
Abstract:
Cataclysmic damage to telecommunication infrastructures, from power grids to satellites, is a global concern. Natural disasters, such as hurricanes, tsunamis, floods, mud slides, and tornadoes have impacted telecommunication services while costing millions of dollars in damages and loss of business. Geomagnetic storms, specifically coronal mass ejections, have the same risk of imposing catastrophic devastation as other natural disasters. With increases in data availability, accurate predictions can be made using sophisticated ensemble modeling schemes. In this work, one such scheme, referred to as stacked generalization, is used to predict a geomagnetic storm index value associated with 2,811 coronal mass ejection events that occurred between 1996 and 2014. To increase lead time, two rounds (stages) of stacked generalization using data relevant to a coronal mass ejection's life span are executed. Results show that for this dataset, stacked generalization performs significantly better than using a single model in both stages for the most important error metrics. In addition, overall variable importance scores for each predictor variable can be calculated from this ensemble strategy. Utilizing these importance scores can help aid telecommunication researchers in studying the significant drivers of geomagnetic storms while also maintaining predictive accuracy.
Pages: 11 to 21
Copyright: Copyright (c) to authors, 2017. Used with permission.
Publication date: June 30, 2017
Published in: journal
ISSN: 1942-2601