Table 2.

Evaluation results on the TITAN and TRANCOS datasets comparing the proposed model with state-of-the-art video-based traffic analysis methods. All models are assessed using Accuracy, Precision, F1 Score, and AUC. The proposed method shows superior performance in noisy and sparse data conditions.

Model	TITAN dataset				TRANCOS dataset
	Accuracy	Precision	F1 Score	AUC	Accuracy	Precision	F1 Score	AUC
SlowFast (44)	84.31 ± 0.03	80.14 ± 0.02	81.27 ± 0.03	85.46 ± 0.02	82.40 ± 0.03	79.88 ± 0.03	80.52 ± 0.02	84.73 ± 0.03
TSN (45)	82.93 ± 0.02	79.77 ± 0.03	80.05 ± 0.02	83.91 ± 0.03	83.67 ± 0.02	81.22 ± 0.03	80.63 ± 0.03	82.98 ± 0.02
VideoMAE (46)	85.76 ± 0.02	83.10 ± 0.02	83.92 ± 0.03	86.79 ± 0.03	84.88 ± 0.03	82.67 ± 0.02	83.19 ± 0.03	85.23 ± 0.02
TimeSformer (47)	83.12 ± 0.03	80.55 ± 0.03	81.78 ± 0.02	84.21 ± 0.02	85.19 ± 0.02	80.46 ± 0.03	82.75 ± 0.02	83.67 ± 0.03
I3D (48)	86.09 ± 0.03	81.73 ± 0.02	84.17 ± 0.03	87.12 ± 0.02	85.42 ± 0.03	83.33 ± 0.02	84.03 ± 0.02	86.88 ± 0.02
X3D (49)	84.88 ± 0.02	82.56 ± 0.03	82.81 ± 0.02	85.87 ± 0.03	83.55 ± 0.02	80.78 ± 0.03	82.04 ± 0.03	84.21 ± 0.02
Ours	89.42 ±0.02	86.15 ±0.03	87.63 ±0.02	90.33 ±0.02	88.91 ±0.03	85.92 ±0.02	87.18 ±0.02	89.87 ±0.03