Scribd
Upload
52 views

Sapient Problem Statement

This document outlines a challenge to predict loan repayment ability using machine learning models. Participants are given training and test datasets containing loan applicant information and repayment labels for the training set. The goal is to predict repayment probabilities for applicants in the test set. A scoring metric of ROC AUC will be used to evaluate predictions against true targets. Participants are asked to submit source code, an architectural diagram, and a report discussing exploratory data analysis and feature engineering. Scalability and fault tolerance must be considered in the proposed system architecture.

Uploaded by

somil dua
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
52 views

Sapient Problem Statement

This document outlines a challenge to predict loan repayment ability using machine learning models. Participants are given training and test datasets containing loan applicant information and repayment labels for the training set. The goal is to predict repayment probabilities for applicants in the test set. A scoring metric of ROC AUC will be used to evaluate predictions against true targets. Participants are asked to submit source code, an architectural diagram, and a report discussing exploratory data analysis and feature engineering. Scalability and fault tolerance must be considered in the proposed system architecture.

Uploaded by

somil dua
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Financing the un-financed (Social financial lending)

Financing or loans are essential for any community, today Banks are desperate
to provide loans to individuals having good tracking financial/credit history. But
many people are not part of this system, they don't have trackable financial
history and thus Banks are reluctant to offer loans to them. And these people are
often exploited by local untrustworthy lenders

Idea at Sapient AI hub is to predict borrowers' repayment capability by using


other easily available dataset like telco and transactional information.

This challenge is to predict loan pay-ability using various statistical and


machine learning methods

Data is broken into two files for training (with TARGET) and testing (without
TARGET).
Static data for all applications. One row represents one loan in data sample.
Data Set  Download Data Set
File Name Description Format
application_train.csv Train Data Set csv
Sample
sample_submission.csv csv
Submission
application_test.csv Test Data Set csv

Data Dictionary
Here's a brief version of what you'll find in the data description file.
Variable Description
SK_ID_CURR applicant_id
probability of applicant paying
TARGET
back
Submission
Data pre-processing:

1. data in application_train.csv may also be corrupt (some of mandatory


fields are missing), filter out those records
2. Since this data was collected from multiple sources hence it may contain
duplicate records, please clean up duplicate records.
Designing and Architectural considerations:

1. Propose an end to end design from collecting uses' data to processing and
the finally accessing output in real time fashion.
2. This application is expected to process huge number of applications in
real time, hence think about scalability.
3. This application is also backbone of entire loan process, hence it's very
important to keep it up all time and think about fault-tolerance.
Expected output/submission:

 You have to predict weather a borrower's is capable of paying loan back.


Your code should generate output with 2 columns – applicant_id and
probability of applicant paying back.
SK_ID_CURR TARGET
100001 0.5
100005 0.5
100013 0.5
100028 0.5
100038 0.5
100042 0.5
100057 0.5
100065 0.5
100066 0.5
100067 0.5
100074 0.5
100090 0.5
100091 0.5
 Complete solution

This should contain the following:

1. A source file containing bug-free code


2. An architectural diagram proposing end to end solution
3. PDF file containing - EDA

Evaluation Metric
1. Your output file will be evaluated on area under the ROC curve between
the predicted probability and the observed target [45 % weightage]
2. 20 % marks will be awarded on code quality
3. 30 % marks will be awarded for feature engineering and EDA
4. 5% bonus marks will be awarded if problem is solved using any big data
technology (like pyspark)

You might also like