LASSO-type penalization in the framework of generalized additive models for location, scale and shape

Groll, Andreas; Hambuckers, Julien; Kneib, Thomas; Umlauf, Nicholaus

doi:10.1016/j.csda.2019.06.005

Cited by 33 publications

(33 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Future work also includes the extension to high-dimensional settings, following for instance Mayr et al (2012) where the problem of variable selection is considered. An alternative strategy for variable selection is developed in Hambuckers et al (2018) and Groll et al (2018) using L 1type of penalties.…”

Section: Discussionmentioning

confidence: 99%

Robust fitting for generalized additive models for location, scale and shape

et al. 2021

View full text Add to dashboard Cite

The validity of estimation and smoothing parameter selection for the wide class of generalized additive models for location, scale and shape (GAMLSS) relies on the correct specification of a likelihood function. Deviations from such assumption are known to mislead any likelihood-based inference and can hinder penalization schemes meant to ensure some degree of smoothness for nonlinear effects. We propose a general approach to achieve robustness in fitting GAMLSSs by limiting the contribution of observations with low log-likelihood values. Robust selection of the smoothing parameters can be carried out either by minimizing information criteria that naturally arise from the robustified likelihood or via an extended Fellner–Schall method. The latter allows for automatic smoothing parameter selection and is particularly advantageous in applications with multiple smoothing parameters. We also address the challenge of tuning robust estimators for models with nonlinear effects by proposing a novel median downweighting proportion criterion. This enables a fair comparison with existing robust estimators for the special case of generalized additive models, where our estimator competes favorably. The overall good performance of our proposal is illustrated by further simulations in the GAMLSS setting and by an application to functional magnetic resonance brain imaging using bivariate smoothing splines.

show abstract

Section: Discussionmentioning

confidence: 99%

Robust fitting for generalized additive models for location, scale and shape

et al. 2021

View full text Add to dashboard Cite

show abstract

“…GAMLSS model was extended by Rigby and Stasinopoulos based on LMS in 2004, and it contains four parameters: μ, σ, ν and τ. Although the model has only been in existence for 17 years, its theoretical system has been gradually improved and developed rapidly due to its powerful func-tions (18), and it has been widely used in the construction of percentile curve of time series data and the formulation of relevant standards (19). In this study, there were different degrees of "skewness" and "kurtosis" in the original data of growth and development indicators of boys and girls aged 3-6 years, which is consistent with the modeling conditions of GAMLSS model.…”

Section: Gamlss Model Has Significant Advantagesmentioning

confidence: 99%

Chinese Preschool Children Growth Reference Standard and Related Curve: Based on GAMLSS Approach

Zhang¹,

Jiang²,

Tu³

2021

ijph

View full text Add to dashboard Cite

Background: The current national growth and development standard of preschool children in China was formulated in 2003, which has many deficiencies. It is necessary to construct more scientific percentile curve and growth reference standards in order to evaluate more effectively the growth, development and health status of Chinese children. Methods: Based on the physical and health data of 31 provinces in China measured in 2010 and 2014, the GAMLSS model was used to construct the growth reference standard and correlation curve. Results: We obtained growth reference standards for percentile curve and Z-score curve of height-for-age, sitting height-for-age, Weight-for-age, Chest circumference-for-age of Chinese preschool children. The C50 percentile of all indicators showed an obvious increasing trend with aged 3.0 to 6.5. Such as, the height of boys and girls increased by 21.1cm and 20.3cm respectively, the sitting height boys and girls increased by 10.3cm and 10.1cm respectively, the weight of boys and girls increased by 7.1 kg and 6.3 kg respectively, the Chest circumference of boys and girls increased by 6cm and 5.2 cm respectively. Conclusion: The children's growth and development charts provided in this study provide effective monitoring and personalized evaluation tools for the growth and development assessment of preschool children, as well as for the reduction of malnutrition, prevention and control of childhood obesity. It is recommended to be used in some areas such as child health, medical treatment and public health.

show abstract

“…GAMLSS have been used in several fields including ML (Hofner et al, 2014;Groll et al, 2019;Mayr et al, 2012), ecology (Smith et al, 2019), survival analysis (De Castro et al, 2010); clinical management of hearing loss (Hu et al, 2015), insurance (Gilchrist et al, 2009), real state appraisal of land lots (Florencio et al, 2012), among others. Softwarewise, GAMLSS is implemented in R through the gamlss package (Stasinopoulos et al, 2007(Stasinopoulos et al, , 2017Rigby et al, 2020).…”

Section: Gamlss As a Distributional Regression Framework For Statistimentioning

confidence: 99%

“…The vector of fixed and/or random-effect parameters are estimated within the GAMLSS framework by maximising the penalised log-likelihood and this can be accomplished by using fast backfitting algorithms and resampling procedures (Rigby and Stasinopoulos (2005); Groll et al (2019); Mayr et al (2012)). Model selection is performed by finding the lowest global deviance GD = −2 (θ ), where (θ ) is the penalised log-likelihood.…”

Section: Gamlss As a Distributional Regression Framework For Statistimentioning

confidence: 99%

“…with x β being a linear predictor. In order to compare two nested competing GAMLSS models M 0 and M 1 based on equation 3, that is, when one model can be obtained from the others by imposing parametric restrictions, the global deviance or a LASSO approach (Groll et al (2019)) can be used to penalise overfittings and select the best model. When comparing two non-nested GAMLSS models (including models with smoothing terms (Hastie and Tibshirani (1990))), the generalised Akaike information criterion (Akaike (1974)) and the J and MJ tests (Davidson and MacKinnon, 1981;Godfrey, 2011;McAleer, 1995) can be used (Cribari-Neto and Lucena (2017)).…”

Section: Gamlss As a Distributional Regression Framework For Statistimentioning

confidence: 99%

See 1 more Smart Citation

Distributional regression analysis of learning analytics and educational data

Marmolejo‐Ramos¹,

Tejo²,

Brabec³

et al. 2020

Preprint

View full text Add to dashboard Cite

The advent of technological developments is allowing gathering large amounts of data in several research fields. Learning analytics/educational data mining (LA/EDM) has access to big observational unstructured data captured from educational settings and relies mostly on unsupervised machine learning algorithms to make sense of such type of data. Generalised additive models of location, scale and shape (GAMLSS) are supervised statistical learning approaches that allow modelling all the parameters of the distribution of the response variable w.r.t. the explanatory variables. This article briefly introduces the power and flexibility of GAMLSS to the LA/EDM community in order to prompt a distributional and interpretable statistical learning of data.

show abstract

LASSO-type penalization in the framework of generalized additive models for location, scale and shape

Cited by 33 publications

References 31 publications

Robust fitting for generalized additive models for location, scale and shape

Robust fitting for generalized additive models for location, scale and shape

Chinese Preschool Children Growth Reference Standard and Related Curve: Based on GAMLSS Approach

Distributional regression analysis of learning analytics and educational data

Contact Info

Product

Resources

About