Outlier Removal in Model-Based Missing Value Imputation for Medical Datasets

Table 2

The best imputation process over each dataset.

Dataset	Missing rate
	10%	20%	30%	40%	50%
Categorical datasets
Lymphography	GA + SVM	IB3 + KNNI	DROP3 + SVM	IB3 + KNNI	DROP3 + KNNI
Nursery	KNNI	GA + MLP	IB3 + MLP	MLP	MLP
Promoters	DROP3 + MLP	IB3/DROP3 + KNNI	IB3/DROP3 + MLP	IB3/DROP3 + SVM	IB3/DROP3 + SVM
SPECT	KNNI	MLP	MLP	MLP	KNNI
Numerical datasets
Blood	GA + KNNI	GA + MLP	GA + MLP	GA + KNNI	DROP3 + MLP
Breast cancer	IB3 + SVM	IB3 + MLP	IB3 + SVM	IB3 + SVM	IB3 + SVM
E. coli	IB3 + KNNI	IB3 + KNNI	IB3 + KNNI	IB3 + KNNI	IB3 + KNNI
Pima	IB3/DROP3/GA + KNNI/MLP/SVM	IB3 + MLP	IB3 + KNNI/MLP; DROP3 + KNNI; GA + MLP	IB3 + KNNI/MLP	IB3 + MLP
Yeast	IB3 + SVM	IB3 + KNNI	IB3 + SVM	IB3 + SVM	GA + SVM
Mixed datasets
Abalone	IB3 + SVM	GA + MLP	GA + MLP	IB3 + SVM	GA + MLP
Acute	GA + SVM; KNNI/MLP/SVM	MLP/SVM	KNNI/SVM	SVM	MLP
Contraceptive	KNNI	SVM	SVM	IB3 + SVM	MLP
Liver_disorders	IB3 + KNNI	IB3 + KNNI	IB3 + KNNI	IB3/GA + SVM	IB3 + KNNI
Statlog	IB3 + KNNI/MLP	IB3 + KNNI/MLP/SVM; GA + MLP	GA + SVM	IB3 + SVM; GA + MLP/SVM	IB3 + MLP
Statlog_German	IB3/DROP3/GA + KNNI/MLP/SVM	IB3/DROP3/GA + KNNI/MLP/SVM	IB3/DROP3/GA + KNNI/MLP/SVM	IB3/DROP3/GA + KNNI/MLP/SVM	IB3/DROP3/GA + KNNI/MLP/SVM