| Transaction Purpose Classification System | |
| A machine learning system that classifies financial transactions based on their purpose text using multiple algorithms including Naive Bayes, Logistic Regression, and Support Vector Machines. | |
| 🌟 Features | |
| Multiple ML Models: Compare performance across different algorithms | |
| Text Preprocessing: Advanced text cleaning with NLTK | |
| Interactive Web Interface: Built with Streamlit | |
| Real-time Classification: Classify new transactions instantly | |
| Model Comparison: Detailed analysis of model performance | |
| LLM Integration Guide: Conceptual approach for transformer-based models | |
| 🚀 Live Demo | |
| Visit the live demo on Hugging Face Spaces: [Your Space URL] | |
| 📊 Model Performance | |
| The system trains and compares three different models: | |
| Naive Bayes: Fast and effective for text classification | |
| Logistic Regression: Good baseline with interpretable results | |
| Support Vector Machine: Often achieves high accuracy | |
| 🛠️ Local Development | |
| Prerequisites | |
| Python 3.8+ | |
| pip | |
| Installation | |
| Clone the repository: | |
| bashgit clone https://github.com/yourusername/transaction-classification.git | |
| cd transaction-classification | |
| Install dependencies: | |
| bashpip install -r requirements.txt | |
| Run the application: | |
| bashstreamlit run app.py | |
| Open your browser and go to http://localhost:8501 | |
| 📁 Project Structure | |
| transaction-classification/ | |
| ├── app.py # Main Streamlit application | |
| ├── requirements.txt # Python dependencies | |
| ├── README.md # Project documentation | |
| └── .gitignore # Git ignore file | |
| 🔧 How It Works | |
| 1. Data Preprocessing | |
| The system preprocesses transaction text by: | |
| Converting to lowercase | |
| Removing punctuation and digits | |
| Removing stopwords | |
| Lemmatizing words | |
| Filtering short words | |
| 2. Feature Extraction | |
| Uses TF-IDF (Term Frequency-Inverse Document Frequency) vectorization to convert text into numerical features suitable for machine learning. | |
| 3. Model Training | |
| Three models are trained and compared: | |
| Naive Bayes: Probabilistic classifier based on Bayes' theorem | |
| Logistic Regression: Linear model for classification | |
| SVM: Support Vector Machine for high-dimensional data | |
| 4. Classification | |
| The best-performing model is used to classify new transactions into categories like: | |
| Rent | |
| Groceries | |
| Utilities | |
| Subscriptions | |
| Transportation | |
| Dining | |
| Shopping | |
| Healthcare | |
| Fitness | |
| 🤖 Large Language Model Approach | |
| Conceptual Implementation | |
| For improved performance, the system could be enhanced using transformer models: | |
| pythonfrom transformers import AutoTokenizer, AutoModelForSequenceClassification | |
| # Load pre-trained BERT model | |
| tokenizer = AutoTokenizer.from_pretrained('bert-base-uncased') | |
| model = AutoModelForSequenceClassification.from_pretrained( | |
| 'bert-base-uncased', | |
| num_labels=num_classes | |
| ) | |
| # Fine-tune on transaction data | |
| # (See full implementation in the app) | |
| Benefits of LLM Approach | |
| Better Context Understanding: Captures semantic meaning | |
| Higher Accuracy: State-of-the-art performance | |
| Transfer Learning: Leverages pre-trained knowledge | |
| Robust to Variations: Handles different phrasings better | |
| Implementation Considerations | |
| Computational Requirements: Needs GPU for training | |
| Training Time: Longer than traditional ML | |
| Model Size: Larger deployment footprint | |
| Complexity: More complex pipeline | |
| 📈 Model Evaluation | |
| Models are evaluated using: | |
| Accuracy: Overall correctness | |
| Precision: Correct positive predictions | |
| Recall: Ability to find all positive cases | |
| F1-Score: Harmonic mean of precision and recall | |
| 🔄 API Usage | |
| While the main interface is web-based, the core functionality can be adapted for API usage: | |
| python# Example classification | |
| def classify_transaction(purpose_text): | |
| cleaned_text = preprocess_text(purpose_text) | |
| vectorized = vectorizer.transform([cleaned_text]) | |
| prediction = model.predict(vectorized)[0] | |
| return prediction | |
| # Usage | |
| result = classify_transaction("Monthly apartment rent payment") | |
| print(f"Predicted category: {result}") | |
| 🚀 Deployment to Hugging Face Spaces | |
| Step 1: Create Space | |
| Go to Hugging Face Spaces | |
| Click "Create new Space" | |
| Choose "Streamlit" as the SDK | |
| Set space name and visibility | |
| Step 2: Upload Files | |
| Upload these files to your space: | |
| app.py | |
| requirements.txt | |
| README.md | |
| Step 3: Configuration | |
| The space will automatically detect the Streamlit app and deploy it. | |
| Step 4: Access | |
| Your app will be available at: https://huggingface.co/spaces/yourusername/yourspacename | |
| 📝 Sample Data | |
| The system includes sample transaction data covering common categories: | |
| Purpose TextTransaction Type"Monthly apartment rent payment"rent"Grocery shopping at walmart"groceries"Electric bill payment"utilities"Netflix monthly subscription"subscription"Gas station fuel"transportation | |
| 🤝 Contributing | |
| Fork the repository | |
| Create a feature branch | |
| Make your changes | |
| Add tests if applicable | |
| Submit a pull request | |
| 📄 License | |
| This project is licensed under the MIT License - see the LICENSE file for details. | |
| 🔮 Future Enhancements | |
| Add more transaction categories | |
| Implement ensemble methods | |
| Add confidence scoring | |
| Include data upload functionality | |
| Add model retraining capability | |
| Implement A/B testing framework | |
| Add logging and monitoring |