Table 7.

The performance of baseline and different FluSense sensor-based models with different feature subset for total daily flu test count and total daily test positive count prediction. The performance was measured with respect to Pearson correlation coefficient (ρ) and Root Mean Squared Error (RMSE) with Leave-One-Day-Out cross-validation experiment.

		Leave-One-Day-Out
		Total Test		Total Positive
Model	Feature	ρ	RMSE	ρ	RMSE
Baseline	Baseline ∈ {dayType, isHoliday}	0.42	5.02	0.19	2.28
Linear Regression	Baseline + top 3	0.53	4.58	0.33	2.07
Ridge Regression	Baseline + top 3	0.53	4.57	0.33	2.07
Poisson Regression	Baseline + top 3	0.43	5.48	0.25	2.24
Gradient Boosted Tree	Baseline + top 3	0.58	4.44	0.45	2.01
Random Forest	Baseline + top 3	0.65	4.28	0.61	1.68
Random Forest	Baseline + tac	0.59	4.34	0.40	2.00
Random Forest	Baseline + csr	0.58	4.39	0.38	2.03
Random Forest	Baseline + cbypt	0.56	4.44	0.36	2.05