Assessing imputation techniques for missing data in small and multicollinear datasets: insights from craniofacial...

Area of Science:

• Craniofacial morphology research
• Medical imaging analysis
• Biometric data science

Background:

• Craniofacial morphology analysis is vital for understanding development, dysmorphologies, and surgical planning.
• Incomplete CT scans lead to missing data, potentially biasing results and reducing statistical power.
• Addressing missing data is critical for accurate craniofacial research.

Purpose of the Study:

• To evaluate imputation techniques for missing craniofacial data.
• To identify the optimal method for small, high-dimensional, and correlated datasets.
• To ensure data integrity in morphometric analyses.

Main Methods:

• Compared five imputation techniques: Mean/Median, k-Nearest Neighbors (kNN), Multiple Imputation by Chained Equations (MICE), Random Forest (RF), and Decision Tree.
• Utilized a dataset with 42 craniofacial variables from 32 observations, introducing 20% random missing values.
• Performance was assessed using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Variance Preservation.

Main Results:

• Random Forest (RF) imputation showed superior performance with the lowest RMSE (1.3987) and MAE (0.4902).
• RF imputation achieved good variance preservation (0.8961), retaining dataset variability.
• Multiple Imputation by Chained Equations (MICE) had lower accuracy (RMSE: 3.0869, MAE: 1.1246) but closer variance preservation (1.0580).

Conclusions:

• Selecting appropriate imputation methods is crucial for small, high-dimensional, correlated datasets in craniofacial morphometry.
• Random Forest (RF) imputation is recommended for its balance of accuracy and variance preservation.
• Effective imputation enhances the reliability of craniofacial morphology studies.