In out-of-sample city testing, the model achieved an average MAPE of 6.78%, outperforming the LSTM (7.45%) and GNN (7.22%) baselines by − 0.67 and − 0.44% points, respectively (95% CI: −0.98 to − 0.36 pp / −0.72 to − 0.16 pp; both p < 0.01), with an overall trend accuracy of 80.9%.
← all excerpts
A multi-source behavioral data framework for interpretable urban tourism forecasting.
1
—
—