Tree-based algorithms for missing data imputation

Bárcena, María Jesús Gómez; Tusell, Fernando

doi:10.1007/978-3-642-57678-2_19

Cited by 4 publications

(4 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Furthermore, they produce partitions that can be represented graphically. CART methods have been used for imputation in a number of different scenarios (see Barcena & Tusell, ; Conversano & Cappelli, ; Harrell, ; Vateekul & Sarinnapakorn, , among others). van Buuren () outline the steps in a CART based imputation algorithm.…”

Section: Methodsmentioning

confidence: 99%

“…Therefore, they could still distort the distribution of the observed data. PMM is a hot-deck method that replaces the missing values with observed measurements, rather than conditional means (Andridge & Little, 2010;Ford, 1983 CART methods have been used for imputation in a number of different scenarios (see Barcena & Tusell, 2000;Conversano & Cappelli, 2002;Harrell, 2001;Vateekul & Sarinnapakorn, 2009, among others). van Buuren (2012) outline the steps in a CART based imputation algorithm.…”

Section: Regression-based Imputation Procedures Replace Missing Valuesmentioning

confidence: 99%

See 1 more Smart Citation

Handling missing data in consumer hedonic tests arising from direct scaling

Franczak

Castura²,

Browne

et al. 2016

Journal of Sensory Studies

View full text Add to dashboard Cite

In sensory evaluation, it may be necessary to design experiments that yield incomplete data sets. As such, sensory scientists will need to utilize statistical methods capable of handling data sets with missing values. This article demonstrates the advantages of a model‐based imputation procedure that simultaneously accounts for heterogeneity while imputing. We compare this model‐based approach to the current state‐of‐the‐art imputation procedures using two real data sets that arose from central location tests. These data sets contain missing values by design. In addition, these data sets have two data sets nested within each of them. We use these nested data sets to validate the results. Compared to the considered state‐of‐the‐art imputation procedures, we find evidence that the model‐based approach is able to recover the group structure and key characteristics of the data sets when a high percentage of the data are missing. Practical applications The model‐based imputation procedure presented in this manuscript is used to analyse incomplete multivariate data sets arising from central location tests. It can be used to analyse incomplete mulitvariate data sets where there is correlation among the variables. Examples include data arising from high fatigue studies, just‐about‐right scale, free choice profiling, or from the ideal profile method.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Regression-based Imputation Procedures Replace Missing Valuesmentioning

confidence: 99%

Handling missing data in consumer hedonic tests arising from direct scaling

Franczak

Castura²,

Browne

et al. 2016

Journal of Sensory Studies

View full text Add to dashboard Cite

show abstract

“…Other missing data mechanisms are respectively Missing Completely at Random, when missingness does not depend on observed and/or missing values in the data matrix, and Not Missing at Random, when given the observed data, the distribution of missing data indicator matrix still depends on the missing values. There are several approaches to deal with missingness (Little and Rubin 1987;McKnight, McKnight, Sidani and Figueredo 2007): ignoring and deleting data (i.e., Listwise deletion and Discarding instances), availablecase methods (i.e., pairwise delection), non-model based imputation procedures (i.e., unconditional mean imputation, conditional mean imputation), model-based imputation procedures, either implicit models, founded on implicit assumptions on the proximity between individuals belonging to the data set (i.e., hot deck imputation, cold deck imputation, substitution method, composite methods (Sande 1983;Ford 1983;David, Little, Samuel and Triest 1986)), or explicit models, based on a statistical model to describe the predictive distribution of missing data, i.e., linear regression, logistic regression (Little 1992;Ibrahim 1990;Ibrahim, Lipsitz and Chen 1999), multiple imputation methods (Rubin 1987), EM algorithm (Dempster, Laird and Rubin 1977), distribution free methods such as non parametric regression (Chu and Cheng 1995) and tree-based methods (Lakshminarayan, Harp, Goldman and Samad 1996;Barcena and Tusell 2000;Conversano 2002, 2008;Conversano and Siciliano 2009). …”

Section: Missing Datamentioning

confidence: 99%

Accurate Tree-based Missing Data Imputation and Data Fusion within the Statistical Learning Paradigm

2012

View full text Add to dashboard Cite

Framework of this paper is statistical data editing, specifically how to edit or impute missing or contradictory data and how to merge two independent data sets presenting some lacks of information. Assuming missing at random mechanism, this paper provides an accurate tree-based methodology for both missing data imputation\ud and data fusion that is justified within the Statistical Learning Theory of Vapnik. It considers both an incremental variable imputation method to improve computational\ud efficiency as well as boosted trees to gain in prediction accuracy with respect to other methods. As a result, the best approximation of the structural functional risk (also known as irreducible error) is reached, thus reducing at minimum the generalization (or prediction) error of imputation. Moreover, it is distribution free, it holds independently of the underlying probability law generating missing data values. Performance analysis is discussed considering simulation case studies and real world applications

show abstract

“…Naturally, the machine learning community was also first to employ CART and RF predictions for treating missing values as part of a conditional mean imputation procedure (Bárcena & Tusell, 2000;Conversano & Cappelli, 2002;Conversano & Siciliano, 2009;Creel & Krotki, 2006;D'Ambrosio, Aria, & Siciliano, 2012;Ishwaran, Kogalur, Blackstone, & Lauer, 2008;Stekhoven & Bühlmann, 2011 (Jacobucci, Grimm, & McArdle, 2017;Morgan & Sonquist, 1963;Strobl et al, 2009).…”

Section: Conditional Mean Imputation By Recursive Partitioning Predicmentioning

confidence: 99%

Recursive partitioning imputation versus full information maximum likelihood in finite mixture modeling

Waldman¹,

Ke²

2021

Preprint

View full text Add to dashboard Cite

It is well established that omitting important variables that are related to the propensity for missingness can lead to biased parameter estimates and invalid inference. Nevertheless, researchers conducting a person-centered analysis ubiquitously adopt a full information maximum likelihood (FIML) approach to treat missing data in a manner that assumes the missingness is only related to the observed indicators and is not related to any external variables. Such an assumption is generally considered overly restrictive in the behavioral sciences where the data are observational in nature. At the same time, previous research has discouraged the adoption of multiple imputation to treat missing data in person-centered analyses because traditional imputation models make a single-class assumption and do not reflect the multiple group structure of data with latent subpopulations (Enders & Gottschall, 2011). However, more modern imputation models that rely on recursive partitioning do not impose a single-class structure to the data. Focusing on latent profile analysis, we demonstrate in simulations that in samples of N = 1,200 or greater, recursive partitioning imputation algorithms can effectively incorporate external information from auxiliary variables to attenuate nonresponse bias better than FIML and multivariate normal imputation. Moreover, we find that recursive imputation models lead to confidence intervals with adequate coverage and they better recover posterior class probabilities than alternative missing data strategies. Taken together, our findings point to the promise and potential of multiple imputation in person-centered analyses once remaining methodological gaps around pooling and class enumeration are filled.

show abstract

Tree-based algorithms for missing data imputation

Cited by 4 publications

References 4 publications

Handling missing data in consumer hedonic tests arising from direct scaling

Handling missing data in consumer hedonic tests arising from direct scaling

Accurate Tree-based Missing Data Imputation and Data Fusion within the Statistical Learning Paradigm

Recursive partitioning imputation versus full information maximum likelihood in finite mixture modeling

Contact Info

Product

Resources

About