A Road Network Enhanced Gate Recurrent Unit Model for Gather Prediction in Smart Cities

Table 1

Evaluation of prediction results in terms of Recall@ and AUC on three datasets.

(a)

NPF	Recall@1	Recall@5	Recall@10	Recall@20	AUC
LSTM	0.0428	0.1677	0.2894	0.4300	0.6951
GRU	0.0712	0.2372	0.3421	0.4771	0.7778
RNN	0.0809	0.2580	0.3570	0.4865	0.8018
ST-LSTM	0.0621	0.2438	0.3663	0.5050	0.7976
STGN	0.0762	0.2728	0.3734	0.5061	0.8177
STGRU	0.0920	0.2829	0.3891	0.5231	0.8290

(b)

OPF	Recall@1	Recall@5	Recall@10	Recall@20	AUC
LSTM	0.0383	0.1638	0.2634	0.4375	0.7140
GRU	0.0455	0.1979	0.3007	0.4656	0.7611
RNN	0.0588	0.2258	0.3260	0.4881	0.8324
ST-LSTM	0.0502	0.2107	0.3174	0.4890	0.7817
STGN	0.0633	0.1699	0.2833	0.4920	0.8292
STGRU	0.0681	0.2439	0.3464	0.5116	0.8545

(c)

TPF	Recall@1	Recall@5	Recall@10	Recall@20	AUC
LSTM	0.0752	0.3331	0.4849	0.6428	0.8317
GRU	0.0789	0.3319	0.4793	0.6384	0.8451
RNN	0.0856	0.3634	0.5066	0.6561	0.8645
ST-LSTM	0.0864	0.3699	0.5213	0.6682	0.8727
STGN	0.0900	0.3327	0.5103	0.6672	0.8734
STGRU	0.0933	0.3795	0.5263	0.6730	0.8755