Table 3.

Ensemble weight details and Sensitivity.

Model	Weight	Rationale	Imbalance Handling	Stability (SD over 12 months)
BERT	0.60	Contextual depth (highest CRI)	Weighted loss + SMOTE	0.01
LSTM	0.25	Sequential patterns	Weighted loss + SMOTE	0.02
Naïve Bayes	0.15	Generative probabilistic baseline	Intrinsic class priors	0.01
Ensemble	1.00	Diversity for robustness	Combined thresholding	0.02