A Comparative Evaluation of Missing Data Imputation Techniques and Their Impact on Classification Performance in Healthcare
Abstract
Mental health disorders such as anxiety, depression, psychosis, and substance-related diseases are increasingly being recognized as critical public health problems in Pakistan, and data-driven methods are being used for clinical research and decision-making processes. However, incomplete data recording and reporting practices are often responsible for missing values in mental health data sets, which can significantly affect the reliability of machine learning models. This paper aims to evaluate the effectiveness of multiple imputation by chained equations (MICE), k-nearest neighbors (KNN), and missforest (MF) imputation methods on a mental health data set from Kaggle that includes clinical and behavioral characteristics related to mental health patients. Prior to the training of the model, the preprocessing steps included balancing of the dataset, the creation of artificial unaccounted data, and the encoding of the categories. Reconstruction measures were used to test the performance of the imputation, and a Random Forest Classifier was used to test the performance of the model for predictions. The reconstruction accuracy for MICE was 94.67%, with a high R2 value of 0.93, showing the best performance overall, according to the testing results, compared to the performance of KNN, which provided a slightly better reconstruction accuracy of 95.67%. MICE provided better performance in the field of predictive modeling compared to KNN and MissForest, with the highest classification accuracy of 74.12% and a Macro F1 Score of 0.79. This highlights the need to test the performance of the models of imputation in terms of their effect on the performance of mental health analytics.
Related
- Comparative Analysis of Microgrid Control Techniques using Variance-Based Methods
- Exploring QoS metrics in SDN-based IoUT: A comparative study of 3D topology design techniques
- Comparative Performance Analysis of Optimization Solvers for Linear Support Vector Machines: Impact of Preconditioning and Dimensionality An Empirical Study Using the Benchopt Framework
- Estimation of the Label-Noise Transition Matrix with Performance Guarantees via Selective Classification
- Comparative Classification of DNS-over-HTTPS Traffic on the CIRACIC-DoHBrw-2020 Dataset Using Logistic Regression, Random Forest and XGBoost
BibTeX
@article{faiz2026comparative,
title = {{A Comparative Evaluation of Missing Data Imputation Techniques and Their Impact on Classification Performance in Healthcare}},
author = {Sana Faiz and Areej Fatemah Meghji and Salahuddin Saddar},
journal = {Sukkur IBA Journal of Computing and Mathematical Sciences},
year = {2026},
month = oct,
doi = {10.30537/v9jwqk89},
url = {https://doi.org/10.30537/v9jwqk89},
note = {CC BY-NC},
}
RIS
TY - JOUR
TI - A Comparative Evaluation of Missing Data Imputation Techniques and Their Impact on Classification Performance in Healthcare
AU - Sana Faiz
AU - Areej Fatemah Meghji
AU - Salahuddin Saddar
PY - 2026
DA - 2026/10/26
JO - Sukkur IBA Journal of Computing and Mathematical Sciences
DO - 10.30537/v9jwqk89
UR - https://doi.org/10.30537/v9jwqk89
AB - Mental health disorders such as anxiety, depression, psychosis, and substance-related diseases are increasingly being recognized as critical public health problems in Pakistan, and data-driven methods are being used for clinical research and decision-making processes. However, incomplete data recording and reporting practices are often responsible for missing values in mental health data sets, which can significantly affect the reliability of machine learning models. This paper aims to evaluate the effectiveness of multiple imputation by chained equations (MICE), k-nearest neighbors (KNN), and missforest (MF) imputation methods on a mental health data set from Kaggle that includes clinical and behavioral characteristics related to mental health patients. Prior to the training of the model, the preprocessing steps included balancing of the dataset, the creation of artificial unaccounted data, and the encoding of the categories. Reconstruction measures were used to test the performance of the imputation, and a Random Forest Classifier was used to test the performance of the model for predictions. The reconstruction accuracy for MICE was 94.67%, with a high R2 value of 0.93, showing the best performance overall, according to the testing results, compared to the performance of KNN, which provided a slightly better reconstruction accuracy of 95.67%. MICE provided better performance in the field of predictive modeling compared to KNN and MissForest, with the highest classification accuracy of 74.12% and a Macro F1 Score of 0.79. This highlights the need to test the performance of the models of imputation in terms of their effect on the performance of mental health analytics.
SN - 2520-0755
SN - 2522-3003
ER -