Selecting Sensitive Parameter Subsets in Dynamical Models with Application to Biomechanical System Identification

Ahmed Ramadan, Connor Boss, Jongeun Choi, N. Peter Reeves, Jacek Cholewicki, John M. Popovich, Clark J. Radcliffe

Research output: Contribution to journalArticle

1 Citation (Scopus)

Abstract

Estimating many parameters of biomechanical systems with limited data may achieve good fit but may also increase 95% confidence intervals in parameter estimates. This results in poor identifiability in the estimation problem. Therefore, we propose a novel method to select sensitive biomechanical model parameters that should be estimated, while fixing the remaining parameters to values obtained from preliminary estimation. Our method relies on identifying the parameters to which the measurement output is most sensitive. The proposed method is based on the Fisher information matrix (FIM). It was compared against the nonlinear least absolute shrinkage and selection operator (LASSO) method to guide modelers on the pros and cons of our FIM method. We present an application identifying a biomechanical parametric model of a head position-Tracking task for ten human subjects. Using measured data, our method (1) reduced model complexity by only requiring five out of twelve parameters to be estimated, (2) significantly reduced parameter 95% confidence intervals by up to 89% of the original confidence interval, (3) maintained goodness of fit measured by variance accounted for (VAF) at 82%, (4) reduced computation time, where our FIM method was 164 times faster than the LASSO method, and (5) selected similar sensitive parameters to the LASSO method, where three out of five selected sensitive parameters were shared by FIM and LASSO methods.

Original languageEnglish
Article number74503
JournalJournal of Biomechanical Engineering
Volume140
Issue number7
DOIs
Publication statusPublished - 2018 Jul 1

Fingerprint

Fisher information matrix
Identification (control systems)
Confidence Intervals
Head

All Science Journal Classification (ASJC) codes

  • Biomedical Engineering
  • Physiology (medical)

Cite this

Ramadan, Ahmed ; Boss, Connor ; Choi, Jongeun ; Peter Reeves, N. ; Cholewicki, Jacek ; Popovich, John M. ; Radcliffe, Clark J. / Selecting Sensitive Parameter Subsets in Dynamical Models with Application to Biomechanical System Identification. In: Journal of Biomechanical Engineering. 2018 ; Vol. 140, No. 7.
@article{a1ea9bf486244dd28b17564ac42564f9,
title = "Selecting Sensitive Parameter Subsets in Dynamical Models with Application to Biomechanical System Identification",
abstract = "Estimating many parameters of biomechanical systems with limited data may achieve good fit but may also increase 95{\%} confidence intervals in parameter estimates. This results in poor identifiability in the estimation problem. Therefore, we propose a novel method to select sensitive biomechanical model parameters that should be estimated, while fixing the remaining parameters to values obtained from preliminary estimation. Our method relies on identifying the parameters to which the measurement output is most sensitive. The proposed method is based on the Fisher information matrix (FIM). It was compared against the nonlinear least absolute shrinkage and selection operator (LASSO) method to guide modelers on the pros and cons of our FIM method. We present an application identifying a biomechanical parametric model of a head position-Tracking task for ten human subjects. Using measured data, our method (1) reduced model complexity by only requiring five out of twelve parameters to be estimated, (2) significantly reduced parameter 95{\%} confidence intervals by up to 89{\%} of the original confidence interval, (3) maintained goodness of fit measured by variance accounted for (VAF) at 82{\%}, (4) reduced computation time, where our FIM method was 164 times faster than the LASSO method, and (5) selected similar sensitive parameters to the LASSO method, where three out of five selected sensitive parameters were shared by FIM and LASSO methods.",
author = "Ahmed Ramadan and Connor Boss and Jongeun Choi and {Peter Reeves}, N. and Jacek Cholewicki and Popovich, {John M.} and Radcliffe, {Clark J.}",
year = "2018",
month = "7",
day = "1",
doi = "10.1115/1.4039677",
language = "English",
volume = "140",
journal = "Journal of Biomechanical Engineering",
issn = "0148-0731",
publisher = "American Society of Mechanical Engineers(ASME)",
number = "7",

}

Selecting Sensitive Parameter Subsets in Dynamical Models with Application to Biomechanical System Identification. / Ramadan, Ahmed; Boss, Connor; Choi, Jongeun; Peter Reeves, N.; Cholewicki, Jacek; Popovich, John M.; Radcliffe, Clark J.

In: Journal of Biomechanical Engineering, Vol. 140, No. 7, 74503, 01.07.2018.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Selecting Sensitive Parameter Subsets in Dynamical Models with Application to Biomechanical System Identification

AU - Ramadan, Ahmed

AU - Boss, Connor

AU - Choi, Jongeun

AU - Peter Reeves, N.

AU - Cholewicki, Jacek

AU - Popovich, John M.

AU - Radcliffe, Clark J.

PY - 2018/7/1

Y1 - 2018/7/1

N2 - Estimating many parameters of biomechanical systems with limited data may achieve good fit but may also increase 95% confidence intervals in parameter estimates. This results in poor identifiability in the estimation problem. Therefore, we propose a novel method to select sensitive biomechanical model parameters that should be estimated, while fixing the remaining parameters to values obtained from preliminary estimation. Our method relies on identifying the parameters to which the measurement output is most sensitive. The proposed method is based on the Fisher information matrix (FIM). It was compared against the nonlinear least absolute shrinkage and selection operator (LASSO) method to guide modelers on the pros and cons of our FIM method. We present an application identifying a biomechanical parametric model of a head position-Tracking task for ten human subjects. Using measured data, our method (1) reduced model complexity by only requiring five out of twelve parameters to be estimated, (2) significantly reduced parameter 95% confidence intervals by up to 89% of the original confidence interval, (3) maintained goodness of fit measured by variance accounted for (VAF) at 82%, (4) reduced computation time, where our FIM method was 164 times faster than the LASSO method, and (5) selected similar sensitive parameters to the LASSO method, where three out of five selected sensitive parameters were shared by FIM and LASSO methods.

AB - Estimating many parameters of biomechanical systems with limited data may achieve good fit but may also increase 95% confidence intervals in parameter estimates. This results in poor identifiability in the estimation problem. Therefore, we propose a novel method to select sensitive biomechanical model parameters that should be estimated, while fixing the remaining parameters to values obtained from preliminary estimation. Our method relies on identifying the parameters to which the measurement output is most sensitive. The proposed method is based on the Fisher information matrix (FIM). It was compared against the nonlinear least absolute shrinkage and selection operator (LASSO) method to guide modelers on the pros and cons of our FIM method. We present an application identifying a biomechanical parametric model of a head position-Tracking task for ten human subjects. Using measured data, our method (1) reduced model complexity by only requiring five out of twelve parameters to be estimated, (2) significantly reduced parameter 95% confidence intervals by up to 89% of the original confidence interval, (3) maintained goodness of fit measured by variance accounted for (VAF) at 82%, (4) reduced computation time, where our FIM method was 164 times faster than the LASSO method, and (5) selected similar sensitive parameters to the LASSO method, where three out of five selected sensitive parameters were shared by FIM and LASSO methods.

UR - http://www.scopus.com/inward/record.url?scp=85046998245&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85046998245&partnerID=8YFLogxK

U2 - 10.1115/1.4039677

DO - 10.1115/1.4039677

M3 - Article

VL - 140

JO - Journal of Biomechanical Engineering

JF - Journal of Biomechanical Engineering

SN - 0148-0731

IS - 7

M1 - 74503

ER -