Deep Learning in Cybersecurity: A Hybrid BERT–LSTM Network for SQL Injection Attack Detection

Table 6

The quantity of alerts generated by each distinct model.

Model	Quantity of alerts	False positive rate	False negative rate
BERT-LSTM	1,247	13%	17%
Word2Vec-LSTM	1,202	25%	31%
GRU	1,271	27%	29%
CNN	980	16%	38%
SVM	2,594	62%	25%