Deep learning for prediction of population health costs

Table 2 Performance assessment

	r	\(\rho\)	MAPE	\(r^2\)	CPM
Spendings in last year	0.418	0.551	2403.30	\(-\)0.005	0.191
Mean of previous spendings	0.464	0.547	2078.76	0.200	0.301
Ridge regression	0.514	0.610	2126.03	0.260	0.285
Neural network	0.524	0.631	2013.35	0.264	0.323
Ridge regression (ensemble)	0.517	0.611	2116.67	0.265	0.288
Neural network (ensemble)	0.527	0.632	2004.33	0.266	0.326
Morbi-RSA model (2018)\(^*\)	na	na	2267.60	0.258	0.242
Morbi-RSA full model\(^*\)	na	na	2233.53	0.263	0.253

The best performance for each evaluation criterion is shown in bold
Evaluation of methods using: Pearson’s correlation (r), Spearman’s correlation (\(\rho\)), mean absolute prediction error (MAPE), R squared (\(r^2\)) and Cumming’s Prediction Measure (CPM). Performance for the Morbi-RSA models on a different data set (\(^*\)) where obtained from [4, 15]. Correlation values where not available (na) for these models

ISSN: 1472-6947