Table 2 Random Forest and LSTM performances for trend forecasting: Models were built on Google Trends alone (a, c) and the combination of Google Trends and Twitter (b, d).

Model	Metrics	Up-trend	Down-trend	Macro avg.	Weighted avg.
(a) Google Trends-Confirmed cases
Random forest	Sensitivity	1	0.33	0.44	0.72
	Precision	0.71	1	0.57	0.64
	F1 score	0.83	0.5	0.44	0.64
LSTM	Sensitivity	1	0.5	0.72	0.86
	Precision	0.92	1	0.83	0.88
	F1 score	0.96	0.67	0.75	0.85
(b) Combined-Confirmed cases
Random forest	Sensitivity	0.92	1	0.64	0.78
	Precision	1	0.43	0.48	0.74
	F1 score	0.96	0.6	0.52	0.74
LSTM	Sensitivity	0.96	0.83	0.93	0.94
	Precision	1	1	0.92	0.96
	F1 score	0.98	0.91	0.91	0.95
(c) Google trends-hospitalization
Random forest	Sensitivity	1	0.75	0.58	0.75
	Precision	0.67	1	0.56	0.67
	F1 score	0.8	0.86	0.55	0.69
LSTM	Sensitivity	0.89	0.58	0.82	0.81
	Precision	1	1	0.82	0.91
	F1 score	0.94	0.74	0.77	0.82
(d) Combined-hospitalization
Random forest	Sensitivity	1	1	0.67	0.83
	Precision	0.9	0.75	0.55	0.70
	F1 score	0.95	0.86	0.60	0.76
LSTM	Sensitivity	1	0.92	0.92	0.94
	Precision	0.95	1	0.93	0.95
	F1 score	0.97	0.96	0.92	0.94

All models were tested during the out-of-sample period from end of March to June 2022.

Quick links

Search