Loan Approval Prediction and Recommendation System

Loan approval prediction system considers a lot of features for loan approval like cibil score,applicant income, applicant assets, co applicant income, amount of loan required, loan term. A user friendly web application shall be developed so that users shall know if they are eligible for loan and the areas where he/she can improve(cibil score etc) to avail loan. All types of loan interests for the amount shall be displayed to the users.

Proposed solution:

Proposed system predicts the loan approval status and provide insights to the rejected applicants using Extreme Gradient Boosting (XGB) algorithm. It implements gradient boosting trees algorithm. High execution speed and high model performance are some of its features. System also recommends other banks to the rejected applicants where they could get loan.

Methodology

The system is developed with machine learning algorithm for prediction of the result. XGBoost(ensemble learning) algorithm is utilized for the prediction. Data collected is preprocessed using numpy, pandas and visualization is done using matplotlib and seaborn libraries. All the outliers and irregularities are analyzed and normalized. Model is trained on the train dataset and tested on test data set using scikit-learn. Accuracy of model obtained is good for the system. Front end developed with HTML, CSS and Javascript is integrated with

database. Flask is used for creating the web application. Database SQLite stores the credentials. The following steps are sequential in building the model:

1. Load all the libraries

2. Load the dataset,

3: Data Cleaning & Feature Engineering

4. Tune and Run the model

5. Score the Test Population

The steps in sequence for the system flow are:

1. User Registration

2. Authentication (Email OTP verification)

3. User log-in(authorization)

4. Loan selection and details entry

5. Prediction and email triggering and

6. Admin loan approval tracking.

7. Admin keeps track of loans by accessing the system via login.