I used IBM Watson Studio (SPSS Modeler) to create a Model for predicting the repaid risk of "USA Local Corporations Loans".
This model is to classify borrowers according to their commitment to repaid the loans. This is to decide whether they can get future loans or not.
Dataset Link:
https://catalog.data.gov/dataset/local-development-corporations-loans
Introduction:
Public authorities are required by Section 2800 of Public Authorities Law to submit annual reports to the Authorities Budget Office that include loans data. Local development corporations are required to report information on the projects they support and how those approved projects are financed (either through grants, loans, or bonds). The dataset consists of loans data reported by Local Development Corporations beginning with fiscal years ending in 2011.
About The Dataset:
- This dataset contains 13,513 loans with 18 variables (columns) on each loan (rows), including loan amount, borrower rate (or interest rate), original loan amount, loan term completed, and many others.
To make sure about the data cleaning process on the Modeler, I make a data cleaning on IBM Data Refinery as a separate process.
Also, I used Watson Studio in a separate exploratory analysis for the same dataset. Through this process, I created an exploratory Notebook, Slides Notebook, and HTML presentation file.
Using IBM Watson Studio, SPSS Machine Learning Modeler:
Steps:
- Checking the data quality
- Cleaning and Normalizing Data
- Fixing data (the dataset has high Standard Deviation Values)
- Divide data into two sections: Training and Test Partitions. This done by: Auto Prep
- Partition the data
- Creating more than one model, and compare the accuracy of each one.
- Analyze the more accurate model, save it as PMML, and deploy it.
- Test the model by entering data and predict the target factor.
I got a model with a prediction accuracy of about 98%.
I think that is a good result.
Thank you
------------------------------
Essam Ali
------------------------------
#SPSSModeler