Separating Event Weighting, Predictor Scaling and Validation Design in Liquefaction Settlement Transfer
Geotechnical Science and Engineering Journal · 14.07.2023 · 2023 (1) · 89-114
DOI: 10.68304/gsej2023.a105 ↗Kurz gefasst
Fachartikel in der Zeitschrift Geotechnical Science and Engineering Journal, online erschienen am 14.07.2023. Verfasst von Lewis David Ashwal.
Abstract (Originalsprache)
Prediction errors for liquefaction-induced settlement depend on which earthquakes are represented in training and on how their observations are weighted. The analysis considers 100 settlement cases in ten reconciled earthquake groups. Record-weighted and event-weighted ridge fits are compared using a common training-derived predictor scale, complete earthquake holdout, constant reference predictors, penalty sensitivity and nested event-level selection. A coupled weighting and scaling calculation provides a separate comparison. At a fixed normalized penalty of one, shared-scale record and event fits yield pooled mean absolute errors of 0.09140 and 0.11515 m, respectively. Nested selection yields 0.09707 and 0.12866 m. The event-weighted fixed fit exceeds the pooled errors of training-mean and training-median predictors, although its event-average error is lower than theirs. With training size fixed at 68 observations, repeated row partitions and earthquake holdout produce mean pooled errors of 0.08029 and 0.09162 m. Excluding two incompletely defined variables produces a seven-input record-weighted nested error of 0.08348 m; this is a sensitivity result rather than an independently validated model selection. Equal-event weighting does not consistently improve transfer in the examined calculations. The result is conditional on a small, uneven compilation and the stated model family, and supports explicit reporting of the prediction target, scaling rule and validation unit.
Zitieren
APA
Ashwal, L. D. (2023). Separating Event Weighting, Predictor Scaling and Validation Design in Liquefaction Settlement Transfer. Geotechnical Science and Engineering Journal, 2023(1), 89-114. https://doi.org/10.68304/gsej2023.a105
Vancouver
Ashwal LD. Separating Event Weighting, Predictor Scaling and Validation Design in Liquefaction Settlement Transfer. Geotechnical Science and Engineering Journal. 2023;2023(1):89-114. doi:10.68304/gsej2023.a105
BibTeX
@article{Ashwal2023,
author = {Ashwal, Lewis David},
title = {Separating Event Weighting, Predictor Scaling and Validation Design in Liquefaction Settlement Transfer},
journal = {Geotechnical Science and Engineering Journal},
year = {2023},
volume = {2023},
number = {1},
pages = {89--114},
doi = {10.68304/gsej2023.a105}
}
EUVIMED