Research Article

A LASSO-Based Prediction Model for Child Influenza Epidemics: A Case Study of Shanghai, China

Table 1

Influenza estimation accuracies of different models. The best performance corresponding to each accuracy metric in each time period is highlighted in boldface. The reported RMSE, MAE, and MAPE scores are relative to the absolute error of the naive method, i.e., the ratio of the error of a given method to that of the naive method. The numbers in parentheses denote the absolute errors of the naive method.

MetricWhole period (week 1, 2017–week 20, 2020)Off-seasonRegular flu seasons (week 40 to week 20 of the following year)
2016–201712017–20182018–20192019–2020

RMSE
 LASSO0.8990.8721.0380.9720.9100.838
 ARIMA0.9931.2010.9791.0660.9920.863
 RF1.1670.9751.5761.2811.0071.206
 OLS1.2261.2902.6041.2700.9340.875
 LSTM1.3561.2801.7461.6421.0431.384
 Naive1 (1822.322)1 (1232.904)1 (1496.251)1 (1561.335)1 (2217.054)1 (2546.548)
MAE
 LASSO1.0370.9281.0701.0811.0511.088
 ARIMA1.1351.2951.0511.0821.2170.976
 RF1.2681.0561.6341.3491.1891.329
 OLS1.4611.3762.8821.4401.1481.202
 LSTM1.5281.4811.6391.7601.3621.495
 Naive1 (1158.927)1 (872.169)1 (1124.75)1 (1154.000)1 (1332.273)1 (1523.909)
MAPE
 LASSO1.4121.0010.9991.1361.0742.076
 ARIMA1.2161.2691.0701.0661.2181.288
 RF1.4501.1371.4431.2771.1681.860
 OLS1.8391.4342.6191.4311.1492.357
 LSTM1.5551.5221.4361.9811.2811.566
 Naive1 (7.963)1 (5.489)1 (7.582)1 (6.248)1 (7.135)1 (15.159)
Correlation
 LASSO0.9650.9360.8260.9560.8440.985
 ARIMA0.9570.8930.8600.9410.8290.981
 RF0.9410.9200.7580.9320.8000.963
 OLS0.9330.8760.1450.9160.8380.984
 LSTM0.9160.8620.4020.8920.8030.950
 Naive0.9560.9160.8470.9440.8180.975
Correlation of increment
 LASSO0.4750.8580.3220.2410.2820.871
 ARIMA0.4030.7750.3020.2920.2840.781
 RF0.3680.8690.2680.1380.1170.700
 OLS0.3580.707−0.1980.2690.3020.788
 LSTM0.3570.816−0.2020.0860.4780.582
 Naive0.4590.8640.3530.3120.2730.790

As the start of the estimation time duration is the first week of 2017, the regular flu season of 2016–2017 was considered to be from week 1, 2017, to week 20, 2017.