Method for Incomplete and Imbalanced Data Based on Multivariate Imputation by Chained Equations and Ensemble...

Area of Science:

• Machine Learning
• Data Science
• Biostatistics

Background:

• Classification analysis faces challenges with incomplete and imbalanced data, impacting classifier training and accuracy.
• Healthcare applications demand high accuracy for imputed values, especially in patient assessments.
• Existing methods struggle to simultaneously address data imputation and class imbalance.

Purpose of the Study:

• To develop a novel algorithmic approach to overcome challenges posed by incomplete and imbalanced datasets.
• To enhance the accuracy of imputation values for training high-performance classifiers.
• To improve the classification of physical fitness assessments in patient populations.

Main Methods:

• A combined approach termed MICEEN (Multivariate Imputation by Chained Equations and Ensemble Learning) was developed.
• Multivariate Imputation by Chained Equations (MICE) was used for accurate imputation value generation.
• Ensemble learning was employed with imputed data; missing values were synthetically generated for minority classes to balance data distribution.

Main Results:

• The MICEEN method demonstrated superior performance in handling incomplete and imbalanced data.
• Experimental results confirmed the method's effectiveness on benchmark and real-world datasets.
• The approach showed significant advantages in physical fitness assessments of tumor patients with varying missing data rates.

Conclusions:

• The MICEEN approach effectively addresses simultaneous data imputation and class imbalance issues.
• This method offers a robust solution for improving classifier performance in data-scarce and imbalanced scenarios.
• The findings have significant implications for accurate patient data analysis in healthcare settings.