AllLife Bank Personal Loan

Campaign
Bank Loan (AI/ML)
Date : 20 July 2023

Proprietary content. © Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited.
Contents / Agenda

● Executive Summary

● Business Problem Overview and Solution Approach

Proprietary content. © Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited.
Executive Summary
● AllLife Bank is a US bank that has a growing customer base. Majority of customers
are liability customers and very small numbers of customers who are also borrowers
(asset customers).
● Alllife bank had one marketing campaign to convert the liability customers to
become borrower's client as well which their success ratio is 9%. Which gives a
platform for to increase debt services and earn more interest.
● Alllife bank needs to introduce promotional offer to liability customers to get into
security account, mortgage account and credit cards and personal loan. Which way
they can earn more interest.
● Study the customers profile based on that they have to offer new promotional loans.
● Most of client use online banking so approach them and promote the different
products to signup.

Proprietary content. © Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited.
Business Problem Overview and Solution Approach

● Bank has many depositor (liability) however they have a smaller number of client
are borrowers(assets) clients. Their income comes from interests, so their primary
goal is increasing borrower's number.

● Through online platform introduce promotional offer for credit cards, mortages,
personal loan and security account.

● Accodring to provided data best method is Decision Tree Modelling.

● Study client’s profile and offer them to get personal loan based on their income,
family size and credit card use and mortgages.

Proprietary content. © Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited.
EDA Results

● Age : The min is 23 , max is 67 , Average is 45 Years old - Data seems is slightly fitting a
uniform distribution.
● The min Experience is 0 years, the max is 43 and the mean is approximately 20 years.
Looks uniform distribution again.
● The min Income is 46k, max is 224K and average is 64K - Dataset is right skewed.
● The min credit card avg. is 0k (which can be reflecting customers who do not own credit
cards), max is 10K and average is approx. 1.9K. Few outliers are there.
● Out of 5000 only 1538 have mortgage. Based on 1538 -The mortgage distribution of the
customers under mortgage is right skewed with a minimum value of approx. 99K to max
value of 635K and mean value between USD 180-200K

Proprietary content. © Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited.
EDA Results

● The Family sizes more than 29.4% are size 1, 25.9% are size 2 then 20.2% are size 3 and
24.4% are size 4.
● 41.9% of customers are 1: Undergrad, 28.1% of customers are 2: Graduate, 30.0% of
customers are 3: Advanced/Professional
● Highest count for zip code the first 2 digit is 94.
● More than 4000 (90%) of customers did accept a personal loan.
● More than 4000 (94%) of customers do not have a CD account.
● More than 4000 (89.6%) of customers do not Securities accounts.
● More than 3000 (70%) of customers do not use a credit card issued by a different bank.
● 59.7% of customers use the online banking services

Proprietary content. © Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited.
EDA Results

● There is a very strong linear correlation between Age and Experience.

● There is a slight correlation between Income and CC Avg

● High concentration of customers who accepted a personal loan are observed at:

● Higher Income level (starting at approx.100K and above)

● Higher CCAvg (starting at approx. 3K and above)

● Higher Mortgage value (approx. USD 300K and above)

● Customers with CD Accounts

● Customers who use credit cards issued from other banks

● Families with sizes 3 or 4

● Customers with Education 2(graduate) or 3 (advanced/professional)

● The income level of the client has a high impact on the client decision to accept a
personal loan, The high the income, the more chances the client will accept a personal
loan.

● As the mortgage value increase, the customer is more likely to accept a personal loan

● Family become bigger; clients are more willing to accept personal loans.

Proprietary content. © Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited.
Data Preprocessing

● ID column: dropped at beginning , Experience : Negative values replaced by positive

● No Particular Outlier affect data processing.

● Data will be split into train and test and validation to evaluate the train model.

● Predicting a customer will accept a loan but, the customer would not accept a loan.
Which is loss of resources

● Predicting a customer will not accept a loan but, the customer would have accepted
a loan. – which is loss of opportunity – very Important as we loose potential
customer.(False Negative)

Proprietary content. © Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited.
Model Building

● create functions to calculate different metrics and confusion matrix

● We use The model_performance_classification and

Proprietary content. © Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited.
Model Performance Summary
● Model is overfitting the data as observed from
the confusion matrix the FN & FP are 0%

● DecisionTreeClassifier(class_weight={0: 0.15, 1:
0.85}, random_state=1)

● The top 5 variables are :

1. Income
2. Education-Graduate

3. CCAvg

4. Education- Advanced/Professional
5. Family

Proprietary content. © Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited.
Model Performance Improvement

● The original tree is quite complicated and is overfitting the training data set, hence
pre-pruning and post pruning are to be considered to improve the model
performance.

● Train a decision tree using effective alphas. The last value in ccp_alphas is the alpha
value (0.17) that prunes the whole tree, leaving the tree, clfs[-1], with one node.

● We have tried three different model Decision Tree sklearn, Decision Tree (Pre-
Pruning) and Decision Tree (Post-Pruning) .

Proprietary content. © Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited.
Model Performance Improvement

Training performance comparison:

● Accuracy 1.0 1.0

● Recall 1.0 1.0

● Precision 1.0 1.0

● F1 1.0 1.0

Decision Tree (Post-Pruning)

● Accuracy 1.0

● Recall 1.0

● Precision 1.0

● F1 1.0

Proprietary content. © Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited.
Model Performance Improvement

The training performance comparison for the Decision Tree models on the loan data shows :

All Three models (Decision Tree sklearn, Decision Tree (Pre-Pruning) and Decision Tree (Post-Pruning) )
achieved perfect accuracy, recall, precision, and F1-score, with each metric being 1.0.

This suggests that the models have correctly classified all instances in the training set, achieving 100%
accuracy and capturing all positive cases without making any false positive predictions.

Proprietary content. © Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited.
APPENDIX

Proprietary content. © Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited.
Data Background and Contents

There are 13 variables and 5000 records.

Income min is 8k and max is 224k. And mean is 73k.

Proprietary content. © Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited.
Happy Learning !

17
Proprietary content. © Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited.