Bankruptcy Predictor

This project is a web application that predicts financial returns based on input CSV files. It consists of a FastAPI-based backend for predictions and a Streamlit-based frontend for user interaction. The service uses a pre-trained XGBoost model to make predictions.

Features

• User-friendly Streamlit frontend for uploading CSV files and viewing results.
• FastAPI backend for handling prediction requests.
• Upload a CSV file containing financial data.
• Validate the CSV file for required columns.
• Predict financial returns using the XGBoost model.
• Return predictions in JSON format.

Requirements

• Python 3.10+
• FastAPI
• Streamlit
• Pandas
• XGBoost

Installation

1. Clone the repository:

   git clone https://github.com/ReynardL/Bankruptcy-predictor.git
   cd Bankruptcy-predictor

2. Install the required packages:

   uv sync

Usage

Run both frontend and backend with Docker Compose

docker-compose up

This will start both the FastAPI backend and the Streamlit frontend.

Start the backend manually

uvicorn App.main:app --reload

Start the frontend manually

If you want to start the frontend manually, run:

streamlit run app.py

After starting, open your browser and navigate to http://localhost:8501 to use the web interface, or to http://127.0.0.1:8000 to see the FastAPI welcome message.

You can use the /predict endpoint to upload a CSV file and get predictions via API, or use the Streamlit frontend for a graphical interface.

Website

You can access the deployed website at:
https://app-978501737888.us-central1.run.app/

API Endpoints

• GET /: Returns a welcome message.
• POST /predict: Accepts a CSV file and returns predictions.
• API Documentation

CSV File Format

The CSV file should contain the following columns:

• Quick Ratio
• Fixed Assets to Assets
• Interest-bearing debt interest rate
• Total debt/Total net worth
• Borrowing dependency
• ROA(C) before interest and depreciation before interest
• Continuous Net Profit Growth Rate
• Research and development expense rate
• Allocation rate per person
• Revenue per person
• Cash/Current Liability
• Accounts Receivable Turnover
• Quick Assets/Total Assets
• Total income/Total expense
• Net Value Per Share (B)
• Cash Flow to Equity
• Non-industry income and expenditure/revenue
• After-tax Net Profit Growth Rate
• Inventory Turnover Rate (times)
• Total expense/Assets
• Net Value Growth Rate
• Operating Expense Rate
• Total Asset Growth Rate
• Cash Turnover Rate
• Current Liabilities/Liability
• Interest Expense Ratio
• Operating Profit Growth Rate
• Long-term fund suitability ratio (A)
• Cash Flow Per Share
• Average Collection Days

API Users

• Target users: Banks giving out loans, investors.
• Expected daily request volume: 5 requests per day.
• User requirements: Batch processing.

Diagram

Below is a text diagram explaining how someone interacts with the API service:

1. User uploads a CSV file to the FastAPI server via the /predict endpoint.
2. FastAPI Server receives the CSV file and validates it for the required columns.
3. If the CSV file is valid, the FastAPI Server inputs the data to the XGBoost Model for prediction.
4. The XGBoost Model returns the predictions to the FastAPI Server.
5. The FastAPI Server sends the predictions back to the User in JSON format.

Performance Metrics (API)

• Response time:
  • ~10s between sending API request and receiving predictions (with cold start).
  • ~0.1s between sending API request and receiving predictions (no cold start).
• Memory consumption:
  • Minimum: 0
  • Maximum: 43% of 512 = 224MB
• Cloud monitoring dashboards:
  • Number of expected requests: ~5 daily