Interpretable classifiers for prediction of disability trajectories using a nationwide longitudinal database

BMC Geriatrics

Table 4 Performance of machine learning for two-class task prediction

	Full variables					Selected variables with LASSO
	LR	SVM	RF	ANN	XGBoost	LR	SVM	RF	ANN	XGBoost
Accuracy	0.760	0.776	0.805	0.748	0.810	0.764	0.770	0.784	0.753	0.796
Recall	0.828	0.843	0.859	0.816	0.861	0.831	0.837	0.847	0.821	0.854
Precision	0.824	0.841	0.867	0.813	0.870	0.828	0.833	0.850	0.818	0.860
F1 Score	0.821	0.841	0.862	0.808	0.864	0.823	0.832	0.848	0.813	0.855
Hamming	0.172	0.157	0.141	0.184	0.139	0.169	0.163	0.153	0.179	0.146
Jaccard	0.707	0.738	0.768	0.690	0.772	0.710	0.724	0.748	0.697	0.759
Kappa	0.554	0.564	0.583	0.532	0.586	0.564	0.565	0.560	0.542	0.573

Accuracy, recall, precision, F1 score were all calculated with weighted metrics. Hamming, Jaccard, and Kappa refer to Hamming distance, Jaccard similarity coefficient, and Cohen’s kappa score

ISSN: 1471-2318