⭐📰 TrustCheck – News Authenticity Classifier (ML + NLP)

A complete, production-ready ML + NLP system to classify news as REAL or FAKE using TF-IDF & Logistic Regression

📘 Project Overview :-

TrustCheck is a machine learning–based system designed to evaluate the credibility of news articles. With rising misinformation across the internet, detecting fake news has become critical.

This project builds a lightweight, offline, real-time fake news classifier using:

• TF-IDF vectorization
• Logistic Regression model
• Custom text preprocessing
• Confidence-based predictions
• Interactive Streamlit app

No external APIs. No HuggingFace. No internet dependency. Everything runs locally on your machine.

🎯 Key Features

1. 🚀 Machine Learning & NLP

• Text cleaning (URLs, extra spaces, newline removal)
• TF-IDF word vectorization (uni-grams + bi-grams)
• Logistic Regression classifier with balanced weights
• High accuracy and precise classification
• Model and vectorizer stored using joblib

2.💡 Streamlit Web App

• Clean and modern UI
• Paste any news text → get REAL/FAKE
• Confidence score displayed
• Outputs results inside a DataFrame
• Works fully offline

3. 🧱 Production-Ready

• Structured folders
• Reusable prediction function
• Modular + scalable code
• Ready for deployment

🧠 Methodology

1️⃣ Data Preparation Dataset contains two classes: 1 → Real News 0 → Fake News Both datasets (real.csv & fake.csv) are merged, cleaned, shuffled, and processed.

2️⃣ Text Preprocessing - Includes:

• Removal of URLs
• Removal of newline characters
• Removal of extra whitespaces
• Combining title + article body
• Regex-based text cleaning

3️⃣ Feature Engineering (TF-IDF)

• Maximum features: 10,000
• N-grams: (1,2) : Converts text into numerical vectors representing word importance

4️⃣ Model Training - Model used:

• Logistic Regression
• class_weight="balanced"
• max_iter=3000

Chosen for:

• High performance on text classification
• Fast and interpretable
• Low computational requirements

5️⃣ Evaluation - Metrics generated:

• Accuracy
• Precision
• Recall
• F1-score : The model performs strongly on both classes (Real & Fake).

6️⃣ Saving the Model - Both the classifier and vectorizer are saved using Joblib:

• models/model.joblib
• models/vectorizer.joblib : These are loaded later in the Streamlit app.

🚀 Running the Application

1. Install Dependencies

• pip install -r requirements.txt

2. Run the Streamlit

• streamlit run Streamlit.py A browser window will open with the TrustCheck interface.

🖥️ Streamlit Output Example

Input (Fake News):

• Breaking: NASA confirms aliens visited the White House last night!

Output: Prediction: FAKE Confidence: 97%

Input (Real News):

• WASHINGTON (Reuters) – The U.S. Senate approved a new budget framework on Monday.

Output:

• Prediction: REAL
• Confidence: 99%

🛠️ Technologies Used :

• 🐍 Python
• 📚 Scikit-Learn
• 🔤 TF-IDF Vectorizer
• 📘 Pandas
• 🔧 Joblib
• 🧼 Regex
• 🖥️ Streamlit
• 📓 Jupyter Notebook

📈 Potential Enhancements

• Integrate deep learning models (BERT, RoBERTa)
• Add explainability (LIME/SHAP)
• Improve dataset variety
• Add title-only vs full-text prediction options

👨‍💻 Author

Ayush

Aspiring Data Scientist & Analyst

• 📫 Email: bhanuseenu914@gmail.com
• 🌐 GitHub: https://github.com/ayush13-0
• ℹ️ LinkedIn: www.linkedin.com/in/ayush130

🛡️ License

• This project is licensed under the MIT License.