Customers have to always face the problem of having to go through all procedure of loan issue to get to know an estimate of the loan amount and sometimes the amount change when issued.
By automating the loan amount to the customers. They can have a faster estimate which will lead them to make a faster decision on taking a mortgage from Bank or Finance Company.
Also, predict Loan Amount can help Bank or Finance Company with their marketing campaigns. By predicting Loan Amounts they can enhance the marketing campaigns and use the prediction in their website or app to attract more people to use their Finance Servises.
Dataset is derived from Fannie Mae’s Single-Family Loan Performance Data with all rights reserved by Fannie Mae.
The sample data used in this project is 2020 3rd Quarter with dimension 2,771,993 rows and 108 columns.
The same data used in this project is provided Here
The target is Original UPB The dollar amount of the loan as stated on the note at the time the loan was originated.
Here you can find the columns description for the data.
Through exploring this data I found the following insights:
-
Retail origination channel is of utmost use by the party that delivered the loan to the issuer.
-
The numbers of customers joining Homeready Program in the 3rd Quartier is very low in comparison with customers issuing mortgage but not in Homeready Program.
-
Majority of customers choose to not have Mortgage Insurance.
-
Loan Purpose for the newest loans is either a refinance mortgage or a purchase money mortgage.
-
Most of the customers[borrower or co-borrower] issuing a mortgage are not qualifies as a first-time homebuyer.
The model I created was to predict the Loan amount and here you can see the model performance. Where the MSE was 22,663.648 So the prediction is off only by $22,664. So now the customer can has a prediction for loan amount with considering $22,664 range of mistake.
-
Reduce the number of predictors by found which columns will really enhance model performance.
-
Create more complicated and interactive visualizations.
-
Can perform Hyperparameters tuning to get the best parameters for the model for example GridSearch, Cross-Validation...etc.
RAPIDS libraries are open source, written in Python, and built on Apache Arrow. The software is being developed in partnership with enterprises globally. RAPIDS also focuses on common data preparation tasks for analytics and data science.
RAPIDS also includes support for multi-node, multi-GPU deployments, enabling vastly accelerated processing and training on much larger dataset sizes.
-
Accelerate your run time with minimal code changes and no new tools to learn.
-
Increase ML model accuracy by iterating on models faster and deploying them more frequently.
-
Reduce training time
-
Open source software and supported by NVIDIA.
-
Large data size
- The data has 2,771,993 rows and 108 columns so using CPU will take a long time to import, clean, train, and test the ML model. Using GPU helped work progress faster.
-
Data Cleaning
- This data need a lot of cleaning so using cudf instead of pandas help to reduce preprocessing time.
-
Limited Time
- I worked on this project for 4 days so have faster processing definitely helps.