Evolved ensemble of detectors for gross error detection.

Nguyen, Tien Thanh ; McCall, John; Wilson, Allan; Ochei, Laud; Corbett, Helen; Stockton, Phil

Abstract

In this study, we evolve an ensemble of detectors to check the presence of gross systematic errors on measurement data. We use the Fisher method to combine the output of different detectors and then test the hypothesis about the presence of gross errors based on the combined value. We further develop a detector selection approach in which a subset of detectors is selected for each sample. The selection is conducted by comparing the output of each detector to its associated selection threshold. The thresholds are obtained by minimizing the 0-1 loss function on training data using the Particle Swarm Optimization method. Experiments conducted on a simulated system confirm the advantages of ensemble and evolved ensemble approach.

Citation

NGUYEN, T.T., MCCALL, J., WILSON, A., OCHEI, L., CORBETT, H. and STOCKTON, P. 2020. Evolved ensemble of detectors for gross error detection. In GECCO '20: proceedings of the Genetic and evolutionary computation conference companion (GECCO 2020), 8-12 July 2020, Cancún, Mexico. New York: ACM [online], pages 281-288. Available from: https://doi.org/10.1145/3377929.3389906

• Conference Name: 2020 Genetic and evolutionary computation conference (GECCO 2020)
• Start Date: Jul 8, 2020
• End Date: Jul 12, 2020
• Acceptance Date: Mar 20, 2020
• Online Publication Date: Jul 8, 2020
• Publication Date: Jul 31, 2020
• Deposit Date: May 15, 2020
• Publicly Available Date: May 15, 2020
• Publisher: Association for Computing Machinery
• Pages: 281-288
• Book Title: GECCO '20: proceedings of the 2020 Genetic and evolutionary computation conference companion (GECCO 2020), 8-12 July 2020, Cancun, Mexico
• ISBN: 9781450371278
• DOI: https://doi.org/10.1145/3377929.3389906
• Keywords: Gross error detection; Ensemble method; Particle swarm optimization
• Public URL: https://rgu-repository.worktribe.com/output/905899