Defaulter Predictor: ML Model Deployment with FastAPI & Flask

Project Overview

This project demonstrates a machine learning model for predicting defaulters. It includes components for model training, deployment, and API access using FastAPI and Flask. The project is structured as follows:

• ML_Case_Study.ipynb: Contains the code for training the machine learning model to predict defaulters.
• model.pkl: A serialized file containing the trained model, saved using pickle.
• app.py: FastAPI application that receives user details in JSON format, computes predicted probabilities based on the model, and returns the results.
• main.py: FastAPI application used for initial testing.
• DataFile.py: Renders the JSON file and sends it to app.py.
• Procfile_Local: Configuration for local deployment with Gunicorn and Uvicorn. Use the following content:

web: gunicorn -w 4 -k uvicorn.workers.UvicornWorker app

• Procfile_Public: Configuration for public domain deployment. Use the following content:

web: uvicorn app
--host=0.0.0.0 --port=${PORT:-5000}

• requirements.txt: Lists required libraries to be installed during deployment.
• results.csv: Contains results of the test dataset.

Running the Project

Local Deployment

1. Navigate to the Project Directory: Ensure you are in the project home directory where the model.pkl file is located.

2. Install Required Packages:

pip install pickle uvicorn

3. ** Configure Procfile_local**: Ensure the Procfile_Local contains the following:

web: gunicorn -w 4 -k uvicorn.workers.UvicornWorker app:app

4. ** Run the FastAPI Application:**

uvicorn app:app --reload

5. Access the API: By default, the FastAPI server runs on port 8000. Navigate to:

• http://localhost:8000/docs for the interactive API documentation
• http://localhost:8000/redocs for an alternative API documentation view

Public Deployment

1. Access the Model API: Once deployed publicly, you can access the API at:

• https://defaulters-prediction.herokuapp.com/docs
• https://defaulters-prediction.herokuapp.com/redocs

Test Data

Use the following test data to evaluate the model. This data yields a probability of 0.03 with a threshold set at 0.5, indicating a non-defaulter response:

{
  "account_amount_added_12_24m": 50956,
  "account_days_in_dc_12_24m": 0,
  "account_days_in_rem_12_24m": 77,
  "account_days_in_term_12_24m": 0,
  "account_incoming_debt_vs_paid_0_24m": 0,
  "account_status": 1,
  "age": 28,
  "avg_payment_span_0_12m": 12.5,
  "merchant_category": "Diversified entertainment",
  "merchant_group": "Entertainment",
  "has_paid": 1,
  "max_paid_inv_0_24m": 91980,
  "num_active_div_by_paid_inv_0_12m": 0,
  "num_active_inv": 0,
  "num_arch_dc_0_12m": 0,
  "num_arch_dc_12_24m": 1,
  "num_arch_ok_0_12m": 2,
  "num_arch_ok_12_24m": 7,
  "num_arch_rem_0_12m": 0,
  "num_arch_written_off_0_12m": 0,
  "num_arch_written_off_12_24m": 0,
  "num_unpaid_bills": 0,
  "status_last_archived_0_24m": 1,
  "status_2nd_last_archived_0_24m": 1,
  "status_max_archived_0_24_months": 3,
  "recovery_debt": 0,
  "sum_capital_paid_account_0_12m": 36163,
  "sum_paid_inv_0_12m": 93760,
  "time_hours": 20.3328
}