Home Loan Status Prediction

Project Overview

Business Understanding: Currently processing home loan application is done manually and it takes about 2-3 days to inform the applicant whether loan is approved or not.

Business Objective: Reduce the time taken to inform the applicants about home loan status

Hypothesis: Develop a machine learning model that can accurately predict whether a loan application is approved or rejected with less amount of time

Solution

Constructed a Flask web application with a Home Loan Status Prediction model that leveraged machine learning algorithms to predict loan approvals and rejections and decreased loan approval time by 90%

Dockerized the application, implemented a CI/CD pipeline, and deployed the app on Azure

Dataset

The dataset consists of 12 independant variables and 1 dependant variable

Independant Variables:

• Loan_ID - unique identifier for each loan
• Gender - Gender of the loan applicant
• Education - Education status of the applicant. Whether the applicant is graduated or not
• Self_Employed - Whether the applicant is self employed or not
• Married - Marital status of the loan applicant
• Dependents - Number of dependents for the loan applicant
• Property_Area - Property area type of the applicant
• ApplicantIncome - Monthly gross income of the applicant
• CoapplicantIncome - Monthly gross income of the Coapplicant
• LoanAmount - Loan amount needed by the applicant
• Loan_Amount_Term - Loan amount term required by the applicant (In months)
• Credit_History - Credit history of the applicant

Dependant variable:

• Loan_Status - Indicates if the loan is approved or rejected

Link to data used in the project - https://github.com/annmary25/Home-Loan-Status-Prediction/blob/main/data.csv

Demo

Loan.Approval.mov

Loan.Rejected.mov

Run Project locally

Using github

• Clone the complete project using git clone https://github.com/annmary25/Home-Loan-Status-Prediction.git
• Complete the following steps inside the directory Home-Loan-Status-Prediction
  • Create python3 virtual environment using virtualenv -p python3 {name_of_environment}
  • Activate the environment using source {path_to_virtual_environment}/bin/activate
  • Install the dependencies using pip install -r requirements.txt
  • Finally run the project using python app.py

Using Docker

Dockerhub link: https://hub.docker.com/repository/docker/annmary25/home-loan-status-prediction/general

Project Approach

Data Ingestion

• The dataset is read and split into training and testing.
• The raw dataset, training dataset and test dataset is saved as csv file in artifacts folder.

Data Analysis

• 68 % of loans are approved
• 60% of the applicants income lies between 25000 to 60000
• 45% of the coapplicant income is 0
• 50% of the totalincome lies between 40000 and 75000
• 60% of the loan amount lies in between 10 lakh and 20 lakh
• Maximum loan amount is 70 Lakh
• Loan amount tenure opted by 83% of applicants is 360 months ie. 30 years
• Applicants with credit history has higher chance of getting loan than applicants without credit hitory.
• 80% of applicants whose loan got approved has credit history
• Loan was rejected for 92% of applicants who did not have credit history
• 81% of loan applicants are male but there isnt any considerable difference between the loans approved to men and women
• 65% of loan applicants are married and there is 68% chance that a loan is approved for a married applicant
• 58% loan applicants does not have dependents
• 78% of loan applicants are graduates and the employment status % of graduates and not graduates are the same.
• 80% of people whose loan got approved are graduates
• 85% of loan applicants are not self employed
• The % of loan approved for self employed people and employees is same
• There isn't a considerable difference in the property area however 37% of the property is in semi urban area.
• The income of employess is higher than self employed people

Data Transformation

• A ColumnTransformer Pipeline is created.
• for Numeric Variables first SimpleImputer is applied with strategy mean , then Standard Scaling is performed on numeric data.
• for Categorical Variables SimpleImputer is applied with most frequent strategy, then ordinal encoding performed , after this data is scaled with Standard Scaler.
• This preprocessor is saved as pickle file.

Model Training

• In this phase multiple models were trained and after performing hyperparameter tuning, the best model was chosen.
• This model is saved as pickle file.

Prediction Pipeline

• The pipeline converts the custom data into dataframe and predict the outcome based on the model chosen.

Model deployment

• Constructed a Flask web application with a Home Loan Status Prediction model to predict loan approvals and rejections and decreased loan approval time by 90%
• Dockerized the application, implemented a CI/CD pipeline, and deployed the app on Azure