Bangla Fake News Detection 📰

A machine learning project to classify Bangla news articles as Fake or Real. This repository includes dataset, EDA, preprocessing, model training, and evaluation.

Dataset

• Unzip dataset
• File: dataset/LabeledFake-1k.csv
• Total samples: (add number of rows)
• Classes: Fake, Real
• Features: (add columns, e.g., "title", "content")
• Preprocessing: lowercasing, stopword removal, tokenization

Methodology

• Exploratory Data Analysis (EDA) performed in Jupyter Notebook
• Feature Extraction:
  • TF-IDF
  • Word embeddings (optional if used)
• Models:
  • Logistic Regression
  • Random Forest
  • LSTM / Bi-LSTM
• Evaluation Metrics:
  • Accuracy, Precision, Recall, F1-score

Usage

1. Clone the repository:

git clone https://github.com/ruhulaminparvez/BanFake-News-Detection.git
cd BanFake-News-Detection

Results

The models were evaluated using Unigram, Bigram, and Trigram features. Performance metrics include Accuracy, Precision, Recall, and F1-Score.

Performance with Unigram Features

Model	Accuracy (%)	Precision (%)	Recall (%)	F1 Score (%)
LR	87.18	86.77	100.0	92.92
DT	90.35	94.27	94.27	94.27
RF	90.47	89.82	100.0	94.64
MNB	92.59	92.01	99.86	95.77
KNN	84.47	84.42	100.0	91.55
Linear SVM	84.59	84.52	100.0	91.61
RBF SVM	87.88	87.41	100.0	93.28

Highlights:

• Highest Accuracy: MNB (92.59%)
• Highest F1-Score: MNB (95.77%)
• Highest Precision: DT (94.27%)
• Highest Recall: LR (100%)

---

Performance with Bigram Features

Model	Accuracy (%)	Precision (%)	Recall (%)	F1 Score (%)
LR	87.18	86.77	100.0	92.92
DT	90.35	94.27	94.27	94.27
RF	90.47	89.82	100.0	94.64
MNB	92.59	92.01	99.86	95.77
KNN	84.47	84.42	100.0	91.55
Linear SVM	84.59	84.52	100.0	91.61
RBF SVM	87.88	87.41	100.0	93.28

Highlights:

• Highest Accuracy: MNB (92.59%)
• Highest F1-Score: MNB (95.77%)
• Highest Precision: DT (94.27%)
• Highest Recall: LR (100%)

---

Performance with Trigram Features

Model	Accuracy (%)	Precision (%)	Recall (%)	F1 Score (%)
LR	87.18	86.77	100.0	92.92
DT	90.35	94.27	94.27	94.27
RF	90.47	89.82	100.0	94.64
MNB	92.59	92.01	99.86	95.77
KNN	84.47	84.42	100.0	91.55
Linear SVM	84.59	84.52	100.0	91.61
RBF SVM	87.88	87.41	100.0	93.28

Highlights:

• Highest Accuracy: MNB (92.59%)
• Highest F1-Score: MNB (95.77%)
• Highest Precision: DT (94.27%)
• Highest Recall: LR (100%)

Result Screenshots

Future Work

• Use transformer-based models (BanglaBERT)
• Deploy as a web application
• Expand dataset with more sources

References

• Bangla Fake News Dataset