Using Machine Learning Techniques to Predict Hospital Admission at the Emergency Department

Georgios Feretzakis; George Karlis; Evangelos Loupelis; Dimitris Kalles; Rea Chatzikyriakou; Nikolaos Trakas; Eugenia Karakou; Aikaterini Sakagianni; Lazaros Tzelves; Stavroula Petropoulou; Aikaterini Tika; Ilias Dalainas; Vasileios Kaldis

doi:10.2478/jccm-2022-0003

Figures & Tables

Weighted Average values of F-Measure and ROC Area for all methods (10-fold cross-validation)

Weighted Average values of F-Measure and ROC Area for all methods - ReplaceMissingValues filters (10-fold cross-validation)

Performance results by class of LogitBoost (10-fold cross-validation)

	TP Rate	FP Rate	Precision	Recall	F-Measure	MCC	ROC Area	PRC Area	Admission
	0.489	0.147	0.658	0.489	0.561	0.370	0.774	0.657	Yes
	0.853	0.511	0.742	0.853	0.794	0.370	0.774	0.854	No
Weighted Avg.	0.719	0.377	0.711	0.719	0.708	0.370	0.774	0.782

Performance results by class of Bagging (10-fold cross-validation)

	TP Rate	FP Rate	Precision	Recall	F-Measure	MCC	ROC Area	PRC Area	Admission
	0.471	0.142	0.657	0.471	0.549	0.360	0.764	0.654	Yes
	0.858	0.529	0.737	0.858	0.793	0.360	0.764	0.840	No
Weighted Avg.	0.716	0.387	0.708	0.716	0.703	0.360	0.764	0.772

Performance results by class of Logistic Regression–ReplaceMissingValues filters (10-fold cross-validation)

	TP Rate	FP Rate	Precision	Recall	F-Measure	MCC	ROC Area	PRC Area	Admission
	0.456	0.143	0.649	0.456	0.536	0.345	0.765	0.643	Yes
	0.857	0.544	0.731	0.857	0.789	0.345	0.765	0.841	No
Weighted Avg.	0.710	0.397	0.701	0.710	0.696	0.345	0.765	0.768

Weighted Average values of F-Measure and ROC Area for all methods -ReplaceMissingValues filters (10-fold cross-validation)

	F-Measure	ROC Area
NaiveBayes	0.663	0.741
Logistic Regression	0.696	0.765
Ada boost	0.674	0.731
Logit boost	0.704	0.757
ClassificationViaRegression	0.691	0.758
Random Forest	0.723	0.789
Bagging	0.712	0.775
Multilayer perceptron	0.697	0.740

Performance results by class of ClassificationViaRegression–ReplaceMissingValues filters (10-fold cross-validation)

	TP Rate	FP Rate	Precision	Recall	F-Measure	MCC	ROC Area	PRC Area	Admission
	0.447	0.145	0.641	0.447	0.527	0.334	0.758	0.638	Yes
	0.855	0.553	0.727	0.855	0.786	0.334	0.758	0.837	No
Weighted Avg.	0.705	0.403	0.696	0.705	0.691	0.334	0.758	0.764

Performance results by class of NaiveBayes (10-fold cross-validation)

	TP Rate	FP Rate	Precision	Recall	F-Measure	MCC	ROC Area	PRC Area	Admission
	0.360	0.091	0.696	0.360	0.474	0.330	0.734	0.602	Yes
	0.909	0.640	0.710	0.909	0.797	0.330	0.734	0.811	No
Weighted Avg.	0.708	0.439	0.705	0.708	0.679	0.330	0. 734	0.734

Performance results by class of NaiveBayes –ReplaceMissingValues filters (10-fold cross-validation)

	TP Rate	FP Rate	Precision	Recall	F-Measure	MCC	ROC Area	PRC Area	Admission
	0.329	0.091	0.677	0.329	0.442	0.300	0.741	0.603	Yes
	0.909	0.671	0.700	0.909	0.791	0.300	0.741	0.822	No
Weighted Avg.	0.696	0.458	0.692	0.696	0.663	0.300	0.741	0.742

Performance results by class of Multilayer perceptron (10-fold cross-validation)

	TP Rate	FP Rate	Precision	Recall	F-Measure	MCC	ROC Area	PRC Area	Admission
	0.542	0.190	0.623	0.542	0.580	0.364	0.742	0.622	Yes
	0.810	0.458	0.753	0.810	0.781	0.364	0.742	0.815	No
Weighted Avg.	0.712	0.360	0.705	0.712	0.707	0.364	0.742	0.744

Weighted Average values of F-Measure and ROC Area for all methods (10-fold cross-validation)

	F-Measure	ROC Area
NaiveBayes	0.679	0.734
Logistic Regression	0.697	0.762
Ada boost	0.685	0.753
Logit boost	0.708	0.774
ClassificationViaRegression	0.691	0.760
Random Forest	0.689	0.757
Bagging	0.703	0.764
Multilayer perceptron	0.707	0.742

Performance results by class of Bagging–ReplaceMissingValues filters (10-fold cross-validation)

	TP Rate	FP Rate	Precision	Recall	F-Measure	MCC	ROC Area	PRC Area	Admission
	0.515	0.161	0.649	0.515	0.574	0.375	0.775	0.654	Yes
	0.839	0.485	0.749	0.839	0.791	0.375	0.775	0.852	No
Weighted Avg.	0.720	0.366	0.712	0.720	0.712	0.375	0.775	0.779

Performance results by class of LogitBoost –ReplaceMissingValues filters (10-fold cross-validation)

	TP Rate	FP Rate	Precision	Recall	F-Measure	MCC	ROC Area	PRC Area	Admission
	0.464	0.135	0.666	0.464	0.547	0.364	0.757	0.641	Yes
	0.865	0.536	0.736	0.865	0.795	0.364	0.757	0.837	No
Weighted Avg.	0.718	0.389	0.710	0.718	0.704	0.364	0.757	0.765

Performance results by class of AdaBoost–ReplaceMissingValues filters (10-fold cross-validation)

	TP Rate	FP Rate	Precision	Recall	F-Measure	MCC	ROC Area	PRC Area	Admision
	0.396	0.133	0.634	0.396	0.487	0.302	0.731	0.604	Yes
	0.867	0.604	0.713	0.867	0.782	0.302	0.731	0.815	No
Weighted Avg.	0.694	0.431	0.684	0.694	0.674	0.302	0.731	0.738

Performance results by class of Random Forest–ReplaceMissingValues filters (10-fold cross-validation)

	TP Rate	FP Rate	Precision	Recall	F-Measure	MCC	ROC Area	PRC Area	Admission
	0.540	0.161	0.660	0.540	0.594	0.399	0.789	0.676	Yes
	0.839	0.460	0.759	0.839	0.797	0.399	0.789	0.858	No
Weighted Avg.	0.729	0.350	0.723	0.729	0.723	0.399	0.789	0.791

Performance results by class of Random Forest– ReplaceMissingValues and ClassBalancer filters (10-fold cross-validation)

	TP Rate	FP Rate	Precision	Recall	F-Measure	MCC	ROC Area	PRC Area	Admission
	0.653	0.243	0.729	0.653	0.689	0.412	0.784	0.767	Yes
	0.757	0.347	0.686	0.757	0.720	0.412	0.784	0.783	No
Weighted Avg.	0.705	0.295	0.707	0.705	0.704	0.412	0.784	0.775

Performance results by class of Multilayer perceptron–ReplaceMissingValues filters (10-fold cross-validation)

	TP Rate	FP Rate	Precision	Recall	F-Measure	MCC	ROC Area	PRC Area	Admission
	0.480	0.161	0.633	0.480	0.546	0.343	0.740	0.617	Yes
	0.839	0.520	0.736	0.839	0.784	0.343	0.740	0.808	No
Weighted Avg.	0.707	0.388	0.698	0.707	0.697	0.343	0.740	0.738

Performance results by class of Logistic Regression (10-fold cross-validation)

	TP Rate	FP Rate	Precision	Recall	F-Measure	MCC	ROC Area	PRC Area	Admission
	0.456	0.142	0.650	0.456	0.536	0.346	0.762	0.641	Yes
	0.858	0.544	0.732	0.858	0.790	0.346	0.762	0.838	No
Weighted Avg.	0.711	0.396	0.702	0.711	0.697	0.346	0.762	0.766

Features

Features	Type	Mean	Standard Deviation
CPK	numerical	179.155	1183.877
CREA	numerical	1.06	0.827
CRP	numerical	39.094	71.48
LDH	numerical	222.327	156.343
UREA	numerical	45.651	33.616
aPTT	numerical	34.227	11.443
DDIMER	numerical	1422.899	2522.921
INR	numerical	1.131	0.571
HGB	numerical	12.87	2.13
LYM	numerical	22.085	11.672
NEUT	numerical	69.478	13.083
PLT	numerical	252.467	87.814
WBC	numerical	9.617	5.153
Age	numerical; Integer*	61.175	20.822
Gender	categorical {Male, Female}
ED Unit	categorical {Urology, Pulmonology, Internal Medicine, Otolaryngology, Triage, Cardiology, General Surgery, Opthalmology, Vascular Surgery, Thoracic Surgery}
Ambulance	Categorical {Yes, No}
Admission	Categorical {Yes, No}

Performance results by class of AdaBoost (10-fold cross-validation)

	TP Rate	FP Rate	Precision	Recall	F-Measure	MCC	ROC Area	PRC Area	Admision
	0.423	0.137	0.642	0.423	0.510	0.323	0.753	0.620	Yes
	0.863	0.577	0.721	0.863	0.786	0.323	0.753	0.836	No
Weighted Avg.	0.702	0.415	0.692	0.702	0.685	0.323	0.753	0.757

Performance results by class of ClassificationViaRegression (10-fold cross-validation)

	TP Rate	FP Rate	Precision	Recall	F-Measure	MCC	ROC Area	PRC Area	Admission
	0.447	0.145	0.641	0.447	0.527	0.334	0.760	0.639	Yes
	0.855	0.553	0.727	0.855	0.786	0.334	0.760	0.839	No
Weighted Avg.	0.705	0.403	0.696	0.705	0.691	0.334	0.760	0.766

Performance results by class of Logit Boost– ClassBalancer filter (10-fold cross-validation)

	TP Rate	FP Rate	Precision	Recall	F-Measure	MCC	ROC Area	PRC Area	Admission
	0.685	0.300	0.696	0.685	0.690	0.385	0.773	0.758	Yes
	0.700	0.315	0.690	0.700	0.695	0.385	0.773	0.779	No
Weighted Avg.	0.693	0.307	0.693	0.693	0.693	0.385	0.773	0.769

Performance results by class of Random Forest (10-fold cross-validation)

	TP Rate	FP Rate	Precision	Recall	F-Measure	MCC	ROC Area	PRC Area	Admission
	0.394	0.103	0.689	0.394	0.501	0.345	0.757	0.650	Yes
	0.897	0.606	0.719	0.897	0.798	0.345	0.757	0.832	No
Weighted Avg.	0.713	0.422	0.708	0.713	0.689	0.345	0.757	0.765

Using Machine Learning Techniques to Predict Hospital Admission at the Emergency Department

Figures & Tables

Fig. 1

Fig. 2

Fig. 3

Performance results by class of LogitBoost (10-fold cross-validation)

Performance results by class of Bagging (10-fold cross-validation)

Performance results by class of Logistic Regression–ReplaceMissingValues filters (10-fold cross-validation)

Weighted Average values of F-Measure and ROC Area for all methods -ReplaceMissingValues filters (10-fold cross-validation)

Performance results by class of ClassificationViaRegression–ReplaceMissingValues filters (10-fold cross-validation)

Performance results by class of NaiveBayes (10-fold cross-validation)

Performance results by class of NaiveBayes –ReplaceMissingValues filters (10-fold cross-validation)

Performance results by class of Multilayer perceptron (10-fold cross-validation)

Weighted Average values of F-Measure and ROC Area for all methods (10-fold cross-validation)

Performance results by class of Bagging–ReplaceMissingValues filters (10-fold cross-validation)

Performance results by class of LogitBoost –ReplaceMissingValues filters (10-fold cross-validation)

Performance results by class of AdaBoost–ReplaceMissingValues filters (10-fold cross-validation)

Performance results by class of Random Forest–ReplaceMissingValues filters (10-fold cross-validation)

Performance results by class of Random Forest– ReplaceMissingValues and ClassBalancer filters (10-fold cross-validation)

Performance results by class of Multilayer perceptron–ReplaceMissingValues filters (10-fold cross-validation)

Performance results by class of Logistic Regression (10-fold cross-validation)

Features

Performance results by class of AdaBoost (10-fold cross-validation)

Performance results by class of ClassificationViaRegression (10-fold cross-validation)

Performance results by class of Logit Boost– ClassBalancer filter (10-fold cross-validation)

Performance results by class of Random Forest (10-fold cross-validation)

Paradigm

My account