MAJ853212/xlm_roberta_fake_news_detection

Urdu Fake News Detection using XLM-RoBERTa

A fine-tuned XLM-RoBERTa model for detecting fake news in the Urdu language. Given an Urdu news article, the model classifies it as either Fake or Real.

---

Model Details

Detail	Value
Base Model	xlm-roberta-base
Task	Binary Text Classification
Language	Urdu
Training Epochs	5
Best Epoch	4
Max Token Length	256
Batch Size	8
Learning Rate	2e-5

---

Test Set Performance

Metric	Score
Accuracy	93.70%
Weighted F1	0.9371
Weighted Precision	0.9372
Weighted Recall	0.9370
AUC-ROC	(add your score here)

Per-Class Results

Class	Precision	Recall	F1	Support
Fake (0)	0.8908	0.8984	0.8946	935
Real (1)	0.9568	0.9534	0.9551	2209

---

Dataset

Split	Samples
Train	70%
Validation	15%
Test	15%

Class Distribution:

• Real News: 14,743 samples
• Fake News: 6,231 samples

Class imbalance was handled using weighted CrossEntropyLoss during training.

---

How to Use

from transformers import pipeline

classifier = pipeline(
    "text-classification",
    model="MAJ853212/xlm_roberta_fake_news_detection"
)

result = classifier("حکومت نے ملک میں تعلیمی اصلاحات کا اعلان کر دیا")
print(result)
# Output: [{'label': 'Real', 'score': 0.96}]

---

Project

This model was developed as a Final Year Project (FYP) on the topic of
"Urdu Fake News Detection using Deep Learning".