Welcome, Guest: Register On Nairaland / LOGIN! / Trending / Recent / NewStats: 3,195,381 members, 7,958,055 topics. Date: Wednesday, 25 September 2024 at 08:12 AM |
Nairaland Forum / Science/Technology / A Model Built To Predict Whether A Loan Applicant Will Default Payment.using SAS (854 Views)
Frank Darko, Ghanaian Inventor Who Built Bicycle That Rides On Water Launches / Kelvin Doe, Sierra Leonean Who Built His Own Radio Station At The Age Of 13 / Anambra Teenager Designed And Built This Drone (Video) (2) (3) (4)
(1) (Reply)
A Model Built To Predict Whether A Loan Applicant Will Default Payment.using SAS by Davidifeoluwa(m): 11:01pm On Apr 01, 2016 |
INTRODUCTION While there have been several delinquencies in the responds of loan applicants, The traditional methods of making loan decision is gradually fading out as predictive analytics (application scoring, behavioral scoring and collection scoring) are gradually replacing traditional way ants. In light of this, I want to demonstrate how to predict whether a loan applicant will default or not using: SAS Enterprise Miner. THE MODELLING STRATEGY THE PROJECT PROCESS 1. BUSINESS OBJECTIVE/UNDERSTANDING A Micro finance bank Institution, who offers loan credits to its customers using traditional scoring method. In recent years, the company has given several applicants loans and many of these accepted applicants have defaulted on their loans. Based on the datasets collected from the past applicant, the institution wants me to build a model to predict whether an applicant will default using Predictive Analytics 2. DATA UNDERSTANDING Data Source: The data used contained 13 variables where 12 variables are the predictor such as geographic, demographic and financial variables, the target variable indicate whether a customer defaulted on the loan. The dataset had earlier been converted to a sas dataset, so the data node was dragged into the workflow from the “data source” Data Visualization: In order to understand the data very well, I used “graph explore” so as to investigate each variables across each records some descriptive analysis ( mean, median, Normal distribution, check for missing values). 3. DATA PREPARATION Data Partitioning: “Data partition” node was used to split the data into 3 different smaller set in the following ratio 40:30:30 . The training sample (30%) was used to build the various models, validation sample (30%) was used to fine tune the model built and a test sample (remaining 30%) was used for final assessment Data Transformation: This includes the data preparation process such as dealing with missing value, removing non-normality and applying variable reduction techniques for some models to perform very well, all these were performed using “Transform Variables”“impute” “Variable selection” nodes 4. MODELLING ILLUSTRATION OF MODELS Models: To stage a very effective model, I decided to make use of four data mining techniques; “Regression” “Neural Network” and “Decision Tree” nodes Regression with “imput”: I consider modeling directly after fixing the missing values problem based on “imputed dataset” so as to improve the performance of the regression model Regression2 with “interactive binning”: I considered building another regression model based on “transformed variable” and “interactive Binning” this is expected to have a better performance compared to the first regression model. Decision Tree: Tree models will handle missing values differently than regression models, that’s why its connected directly to “data partition” node directly Model Comparison Neural Network: The final method considered was “Neural network” so as to compare how well a neural network model compares to the regression and the decision tree model 5. MODEL EVALUATION Using “Model Comparison node” The outcome of each of the models were compared in terms of accuracy and prediction power in order to select the one with the best performance. The result showed that Regression2 had the best performance with more 90% accuracy in terms of classification rate Note: some other factors aside from classification accuracy were considered before arriving at the best performing model. Such as lift curve graphs, confusion matrix etc. 6. MODEL DEPLOYMENT FOR PREDICTION Note: Data Mining project is not yet complete until the model built is deployed in an operational environment. Therefore in order to predict the applicants that will default loan payments, the model built was applied on a new datasets with similar variables. The "Score" node scores the new datasets, while the "Reporter" node gives a full report of the whole process and result in a pdf format CONCLUSION AND RECOMMENDATION This model will be able to identify the right applicant, thereby minimizing the likelihood of loan delinquencies among the recipient of loans, because the applicant that will be granted loan will be based on predicted cut-offs by the new score card. Thankew score card. Thanks for reading through… Oladapo Ifeoluwa(2348067096726) ifeoluwa.oladapo@gmail.com https://www./model-built-predict-whether-loan-applicant-default-david-ifeoluwa?trk=hp-feed-article-title-publish 1 Share
|
Re: A Model Built To Predict Whether A Loan Applicant Will Default Payment.using SAS by Flexherbal(m): 11:05pm On Apr 01, 2016 |
This is quite a long piece. Hope some people learn from it. |
(1) (Reply)
---- BITX ---- / Successful People Don't Jog And Sweat On The Streets - Buhari Tells Zuckerberg / Install Microsoft .NET Framework 4.5 Or 3.5.1 On Windows 7
(Go Up)
Sections: politics (1) business autos (1) jobs (1) career education (1) romance computers phones travel sports fashion health religion celebs tv-movies music-radio literature webmasters programming techmarket Links: (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Nairaland - Copyright © 2005 - 2024 Oluwaseun Osewa. All rights reserved. See How To Advertise. 14 |