angang-li / find_lenders

A web app to predict lenders of real-estate deals

Predict Potential Lenders

• Predict Potential Lenders
  • 1. Background
  • 2. Dataset
  • 3. File Descriptions
  • 4. Instructions to run the web application
  • 5. Python version and libraries
  • 6. Results and screenshots

1. Background

This project uses data of real estate loaning deals to solve 3 problems:

• Probalem 1: Use Google Geocoding API to get City and State names based on Zip Code.
• Probalem 2: Classify top 3 lenders who are most likely to provide the loan.
  • Problem 2.1: Build a classification algorithm to predict lenders and their probabilities
  • Problem 2.2: Build a web application that uses the algorithm to predict lenders based on user inputs
• Probalem 3: Find the variables that correlate with higher interest rate.

2. Dataset

The dataset of ~30K real estate loan deals was given by Enodo

Variables list:

• Deal ID, Lender ID, Zip, City* (blank to be filled), State* (blank to be filled), Multifamily Subtype, Built, Units, Original Loan, Note Rate, Loan Term (Original), Appraised Value*, Maturity Date, UPB , Amort DSvc, Orig Amort, IO Period
• Features with * were used to solve Problem 2.
• All features were used to solve Problem 3.

3. File Descriptions

• app

  • template
    • master.html: homepage of web app
    • go.html: classification result page of web app
  • static
    • js
      • app.js: JavaScript file that supports master.html
  • app.py: Flask file that runs web app
  • utils.py: customized functions that support app.py
  • IMPORTANT: in order to run the web app, please add a config.py file that specifies your own Google Geocoding API Key google_api_key = MY-GOOGLE-API-KEY.

• data

  • lender_data.csv: raw lender data to be processed
  • lender_data1.db: lender data processed by part1.ipynb

• model

  • model_rf.pkl: trained classification model that predicts most likely lenders
  • features.pkl: input features of the classification model
  • imputer.pkl: fitted imputer to preprocess data for the classification model
  • scaler.save: fitted scaler to preprocess data for the classification model

• notebook

  • part1.ipynb: notebook that solves Problem 1
  • part2.ipynb: notebook that solves Problem 2.1
  • part3.ipynb: notebook that solves Problem 3
  • IMPORTANT: in order to run the notebooks, please add the following
    • a config.py file that specifies your own Google Geocoding API Key google_api_key = MY-GOOGLE-API-KEY
    • a temp folder at this level

4. Instructions to run the web application

1. Run the following command in the app directory to launch the web app: python app.py.
   
2. Go to http://0.0.0.0:3001/.

5. Python version and libraries

The code was developed using the Anaconda distribution of Python version 3.6. The following dependencies were used.

pandas
numpy
sqlalchemy
sklearn
flask
matplotlib
seaborn
requests
pickle

6. Results and screenshots

• Problem 1: Retrieve city and state based on zip code

  

• Problem 2: Classify top 3 most likely lenders

  • Classifier that predicts top 3 lenders and their probabilities: see part2.ipynb
  • Web app
    • Homepage
      
    • Classification result
      

• Problem 3: Identify variables that correlate with high interest rate

  
  Some of the lenders and loan term correlate strongly with higher interst rate.