An artificial neural network-based critical values for multiple hypothesis testing: data-snooping case

Rofatto, Vinicius Francisco; Matsuoka, Marcelo Tomio; Klein, Ivandro; Bonimani, Maria Luísa Silva; Rodrigues, Bruno Póvoa; Campos, Caio César de; Veronez, Maurício Roberto; Silveira, Luiz Gonzaga da

doi:10.1080/00396265.2021.1968176

Cited by 3 publications

(1 citation statement)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Of course, this requires some computational cost, but nothing prohibitive. One way to save computational costs would be to develop a methodology based on artificial neural networks, as done for example by [44], or some other alternatives, such as [45]. This will be investigated in future work.…”

Section: Contributionsmentioning

confidence: 99%

A New Sequential Statistical Test Procedure from Observation Differences for Geodetic Deformation Analysis

Matsuoka,

Rofatto,

Assunção

et al. 2023

Preprint

View full text Add to dashboard Cite

One of the main challenges in geodetic deformation analysis is to infer whether the geometry of some engineering structures or zones of natural hazards has changed from their initial state or not. The pillar that supports such an analysis is tightly based on the fundamentals of statistical hypothesis testing. The null hypothesis model indicates that no displacement has occurred. It is tested against a class of alternative models, which stipulate different displacement patterns. In this contribution, we present an innovative geodetic displacement detection which integrates combinatorial analysis and likelihood ratio tests into a sequential procedure for the case where the differences between observations of two epochs are in play. This framework is applied and investigated to two test scenarios: a synthetic and a real simulated trilateration network. From a statistical point of view, our approach is rigorous, because the alternative model can identify simultaneously more than one unstable point. In addition, the relationship between the unknown parameters and the observations is always linear, even if the problem manifests itself as non-linear. Consequently, we avoid a potential loss of statistical test power due to the model linearization. In addition, the proposed method controls the false positive rate efficiently. One of the problems that arise here is also related to the selection of the maximum number of points to be considered in the test procedure ( ). Here, we provide an innovative methodology based on rank computation of the design matrices to define , which can even be extended to the problem of outlier detection. Determining avoids the problem of having non-separable models in identifying unstable points. The algorithms and data are available in the repository.

show abstract

Section: Contributionsmentioning

confidence: 99%