Abusive Language Detection on Darkweb, PAN12, and Roman Urdu dataset (5-fold cross-validation)
99.1Average Precision (AP)Hybrid Model
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Hybrid ModelModel category=Hybrid2026.02 | 99.1 | 98.6 | 99.5 | 98.9 | 99.2 | 5,076.8 | 459.78 | |
| BERTModel category=Deep Learning2026.02 | 96.6 | 96.1 | 98.4 | 96.4 | 97.6 | 4,672.8 | 203.6 | |
| LSTM_w2vModel category=Deep Learning, Feature representation=w2v2026.02 | 94 | 90.5 | 96.6 | 92.2 | 93.5 | 233.9 | 1.38 | |
| SVM_w2vModel category=Machine Learning, Feature representation=w2v2026.02 | 93.7 | 89.2 | 96.3 | 91.4 | 93.7 | 340.9 | 42.65 | |
| LR_w2vModel category=Machine Learning, Feature representation=w2v2026.02 | 89.7 | 83.3 | 94.2 | 86.4 | 90.3 | 26.7 | 0.03 | |
| CNN_w2vModel category=Deep Learning, Feature representation=w2v2026.02 | 81.9 | 70.7 | 90 | 75.9 | 82.9 | 192.1 | 1.8 | |
| NB_tf.idfModel category=Machine Learning, Feature representation=tf.idf2026.02 | 69.6 | 89.6 | 89 | 78.3 | 89.2 | 3.7 | 1.75 |