Session
Civil Engineering, Infrastructure and Environment
Description
This paper provides an analysis of the interpretable machine learning (IML) approach applied to various nonlinear dynamic systems. The study focuses on modeling the restoring force by neural networks with two input values: displacement and velocity. A parallel neural network with the proposed architecture, initialized randomly, mirrors the designed model. Both models undergo multiple training and testing cycles on identical datasets to ensure robust statistical validation. The findings demonstrate that interpretable artificial neural networks (ANNs) outperform randomly initialized models in terms of accuracy and result consistency. This research provides valuable insights into the application of IML techniques for advancing the modeling capabilities of nonlinear dynamic systems.
Keywords:
Nonlinear dynamic systems, Interpretable Machine Learning, Artificial Neural Networks, Initialization
Proceedings Editor
Edmond Hajrizi
ISBN
978-9951-982-15-3
Location
UBT Kampus, Lipjan
Start Date
25-10-2024 9:00 AM
End Date
27-10-2024 6:00 PM
DOI
10.33107/ubt-ic.2024.316
Recommended Citation
Morina, Liron; Peja, Rina; and Vranovci, Getoar, "Exploring Interpretable Machine Learning for Modeling Nonlinear Dynamic Systems" (2024). UBT International Conference. 15.
https://knowledgecenter.ubt-uni.net/conference/2024UBTIC/CEIE/15
Included in
Exploring Interpretable Machine Learning for Modeling Nonlinear Dynamic Systems
UBT Kampus, Lipjan
This paper provides an analysis of the interpretable machine learning (IML) approach applied to various nonlinear dynamic systems. The study focuses on modeling the restoring force by neural networks with two input values: displacement and velocity. A parallel neural network with the proposed architecture, initialized randomly, mirrors the designed model. Both models undergo multiple training and testing cycles on identical datasets to ensure robust statistical validation. The findings demonstrate that interpretable artificial neural networks (ANNs) outperform randomly initialized models in terms of accuracy and result consistency. This research provides valuable insights into the application of IML techniques for advancing the modeling capabilities of nonlinear dynamic systems.