🧠 Document Classification using Machine Learning

This project is an intelligent document classification system that uses a range of machine learning models, traditional text vectorization techniques (TF-IDF), and modern embeddings (Word2Vec & Doc2Vec) to accurately classify text documents into predefined categories. Ensemble techniques like hard voting and soft voting are used to improve performance by combining multiple models.

---

📁 Datasets Used

We combined two existing datasets to build a richer and more diverse text classification corpus:

1. News Article Category Dataset
2. Text Document Classification Dataset

These datasets were mapped into unified categories such as:

• News & Current Affairs
• Business & Finance
• Science & Technology
• Arts & Entertainment
• Education & Academia
• Sports

---

🧩 Word Embedding

We used the pre-trained Google News Word2Vec model (GoogleNews-vectors-negative300.bin) for document vectorization:

📥 Download it here: Google News Word2Vec Embeddings

---

🚀 Project Pipeline

1. Data Loading & Preprocessing
2. Category Mapping & Merging Datasets
3. Text Cleaning, Tokenization, and Lemmatization
4. Vectorization:
   • TF-IDF
   • Word2Vec
   • Doc2Vec
5. Oversampling for Class Imbalance (SMOTE, ADASYN, Random Oversampling)
6. Model Training:
   • Naive Bayes
   • SVM
   • Random Forest
   • AdaBoost
   • XGBoost
   • Word2Vec + Logistic Regression
   • Doc2Vec + Logistic Regression
7. Model Evaluation
8. Ensemble Voting (Hard & Soft)
9. Deployment via Streamlit Interface

---

📦 First-Time Setup Instructions

# Clone the repository
git clone https://github.com/Kakarotprince/FileClassification.git
cd FileClassification

# Install dependencies
pip install -r Requirements.txt

⚠️ Important Note:

• On first run, the Google News vectors need to be loaded and vector cache saved.
• Subsequent runs will reuse the saved vectors to save time and memory.

---

📊 Models and Techniques

• Vectorizers: TF-IDF, Word2Vec (Google News), Doc2Vec
• Classifiers: SVM, RandomForest, AdaBoost, XGBoost, Naive Bayes, Logistic Regression
• Imbalanced Data Handling: SMOTE, ADASYN, Random Oversampling
• Evaluation Metrics: Accuracy, Precision, Recall, F1-Score
• Ensembling: Hard Voting, Soft Voting (based on individual model accuracies)

---

🖼️ Streamlit Web Interface

The project includes a user-friendly Streamlit-based UI where users can upload text or files and receive classification results in real-time.

To launch the app:

streamlit run app.py

---

📚 References

• scikit-learn
• Gensim
• Streamlit
• Kaggle: GoogleNews Word2Vec
• Kaggle: News Article Categories
• Kaggle: Text Document Dataset

---