Table 2.

Comparative analysis of data quality improvement strategies in healthcare.

Study	Methods used	Accuracy/AUC	Completeness	Reproducibility/strategy
This study	KNN imputation, anomaly detection (IF, IQR, LOF), PCA, RF	75.3%, AUC 0.83	~100%	MLflow tracking, reproducible pipeline
Chen et al. (2021)	Transfer learning, data quality evaluation, CNN/RNN	~ + 11% accuracy improvement after cleaning	Implicitly improved	Structured pipeline with medical concept normalization
Kale and Pandey (2024)	KNN, SMOTE, clustering-based anomaly detection, multiple classifiers	Accuracy 70% → 91%, F1-score 0.65 → 0.89	Substantial post-imputation	Clear before/after metric reporting
Emery et al. (2024)	Imputation: MICE, hot-deck, log-linear, MICT-timing	65–74% (simulated data)	Trajectory coverage improved	Reproducible in R, used on longitudinal data
Azimi and Pahl (2024)	Anomaly analytics, ML quality metadata exploration	Not quantified; improved ML output reliability	Qualitative evaluation	Anomaly visualization and root cause strategy