Network analytics and machine learning for predicting length of stay in elderly patients with chronic diseases at point of admission

Table 4 Top ten features in tree-based models

XGBoost	RI^a	GBDT	RI	RF	RI
mean of neighbors’ LOS	1	mean of neighbors’ LOS	1	mean of neighbors’ LOS	1
median of historical LOS	0.56	mean of historical LOS	0.74	mean of historical LOS	0.57
max of historical LOS	0.38	last LOS	0.45	median of neighbors’ LOS	0.41
mean of historical LOS	0.28	median of neighbors’ LOS	0.3	median of historical LOS	0.39
LDA-1	0.24	std of neighbors’ LOS	0.25	max of historical LOS	0.16
std of neighbors’ LOS	0.23	last discharge interval	0.21	last LOS	0.14
last LOS	0.21	median of historical LOS	0.19	last discharge interval	0.13
median of neighbors’ LOS	0.2	max of historical LOS	0.19	LDA-1	0.13
last discharge interval	0.17	hospital address	0.16	std of neighbors’ LOS	0.12
LDA-2	0.16	LDA-1	0.13	hospital address	0.1

^aRI is the relative importance of using min–max normalization. The LDA-1 represents the first component after LDA reduction for network features

ISSN: 1472-6947