PlumX Metrics
Embed PlumX Metrics

A Comparison of Imputation Methods for Categorical Data

SSRN, ISSN: 1556-5068
2023
  • 0
    Citations
  • 319
    Usage
  • 0
    Captures
  • 0
    Mentions
  • 0
    Social Media
Metric Options:   Counts1 Year3 Year

Metrics Details

  • Usage
    319
    • Abstract Views
      254
    • Downloads
      65
  • Ratings
    • Download Rank
      702,318

Article Description

Objectives Missing data is commonplace in clinical databases, which are being increasingly used for research. These databases contain mainly categorical variables. The questionable aspect is the best imputation method for categorical data. Materials and methods We utilized data extracted from paper-based maternal health records from Kawempe National Referral Hospital, Uganda. We compared the following imputation methods for categorical data in an empirical analysis: Mode, K-Nearest Neighbors (KNN), Random Forest (RF), Sequential Hot-Deck (SHD), and Multiple Imputation by Chained Equations (MICE). In the first approach to compare the imputation methods, random missing data was injected at varying proportions in the complete dataset (5%-50%). The missing values were imputed by the five imputation methods which were then compared by precision score. In the second approach, the complete dataset was split into training and testing dataset. Random missing data (5%-50%) was then injected into only the training set. Imputation methods were then compared by predictive accuracy of the outcome variable in four classifiers on a single testing set. The consistency of performance of imputation methods was assessed by Kendall’s W test. Results KNN imputation had the highest precision score at levels of missing data (Kendall’s W = 0.853, p = 0.0000842). However, the methods performed differently at all proportions of missing data in the four classifiers. Conclusions KNN imputation is the best method in predicting missing values in categorical variables. There is no universal best imputation method that yields the highest predictive accuracy at all proportions of missing data.

Bibliographic Details

Shaheen M.Z. Memon; Ignace H. Kabano; Robert Wamala

Elsevier BV

Multidisciplinary; Imputation; categorical variables; precision score; single imputation; multiple imputation

Provide Feedback

Have ideas for a new metric? Would you like to see something else here?Let us know