Comparison of Data Imputation Techniques in Handling Missing Data for Health Informatics
Repakula Santhoshi,
Sannella Prabhaker,
Rachel R
Health informatics systems continuously generate enormous volumes of patient-related information from
hospitals, diagnostic laboratories, wearable devices, electronic health records (EHRs), medical imaging
systems, and clinical monitoring platforms. However, these datasets frequently contain missing values
resulting from incomplete patient records, sensor failures, human errors, inconsistent documentation,
communication failures, and data integration issues. Missing data significantly degrades the performance
of predictive analytics, disease diagnosis models, clinical decision support systems, and healthcare
management applications. Therefore, selecting appropriate data imputation techniques has become an
essential preprocessing step for improving the quality and reliability of healthcare analytics. This paper
presents a comprehensive comparison of data imputation techniques for handling missing data in health
informatics. The proposed framework evaluates commonly used statistical, machine learning, and deep
learning-based imputation methods, including Mean Imputation, K-Nearest Neighbor (KNN), Multiple
Imputation by Chained Equations (MICE), Random Forest Imputation, and Autoencoder-Based Deep
Learning Imputation. Performance evaluation is conducted using healthcare datasets containing patient
demographics, laboratory reports, physiological measurements, and clinical records with artificially
introduced missing values. Comparative analysis is performed using imputation accuracy, Root Mean
Square Error (RMSE), Mean Absolute Error (MAE), and computational efficiency. Experimental results
demonstrate that deep learning-based imputation methods provide superior reconstruction accuracy
and preserve underlying data distributions more effectively than traditional statistical approaches. The
proposed comparative framework assists healthcare researchers and practitioners in selecting suitable
imputation techniques for improving predictive model performance and clinical decision-making in
health informatics applications.
