Motivation Letter for ORBA students

First Name: Pankaj

Last Name: Pankaj

A) Motivation
Please specify why the study program ORBA is important for your future
career. (800 characters)

Letter of Motivation - Operations Research and Business Analytics (ORBA).

Hello, my name is Pankaj Khandelwal and I was born and raised in New Delhi – India. I completed my
Bachelor’s degree in Hospitality and Hotel Administration in 2015 from the Institute of Hotel
Management – PUSA(New Delhi). After which I joined a company Foodcraft India Pvt Ltd. which is in
cloud kitchen service and I worked for one year as a shift manager. I worked there in areas of daily
operations activity, vendor management, Inventory management, storekeeping, and MIS reporting.

After working for one year in Foodcraft India Pvt Ltd I left my job to start my own business ‘Wolf of Eat
Street’ in the street food truck industry. I started my venture from available resources from 2016 to 2018
in New Delhi. I operated my food-truck business for two years but due to some uncertainty and legal
issues, I had to shut it down in late 2018. This startup helped me to develop key skills like- Entrepreneur
mindset, business modeling, leadership, strategic decision making and was the turning point that excites
me more about learning the business concepts and strategies. This failed start-up make me realize how
we were doing good in products and innovations but failing in doing business and delivering services.

In 2018 I decided to go to business school to pursue my higher studies and take admission to Fortune
Institute of International Business for completing my Post Graduation Diploma in
Management(PGDM 2019-2021). After enrolling with the PGDM program at FIIB I learned how data and
analytics come into hand in hand for business intelligence, strategic decision making, and SMART
problem-solving approach. This creates my inclination to bend towards analytics and as I had interest
and prior experience working in a core operations job, I choose my major courses as Business
operations and Analytics during my PGDM.

During my PGDM program, I completed my 14 days Social Internship Program with Friends and
Smile NGO and I am still an active volunteer in NGO Friends and Smile. I lead all their events related
operational and planning activities held by the foundation for fundraising.

I did my 3 months corporate internship program with YOMA multinational, where my project was that
company wants to develop a freemium model of HRIS/HCM platform for SMEs. While providing limited
free services for HR operations and using client’s employee data for selling financial services. As the
company was planning to enter the market and trying to design the product according to market needs it
gave me the opportunity to work on market analysis, swot analysis, positioning strategy, secondary
research for the product and helped me gain skills like Writing functional requirements, design thinking,
technical writeup of product specifications.

I got my campus placement and joined Demour Group, currently working as an operations specialist
with client Kuehne-Nagel Integrated logistics team handling Moderna vaccine shipments planning with
Luxemburg team.
 I have chosen the Master’s program of ‘Operations Research and Business Analytics (ORBA)’
at the university of ‘Otto von Guericke University Magdeburg’ due to some specific reasons. The
master’s program is chosen because of its closely related to my previous study, makes me
specialized in this field of study, and is closely aligned with my career domain Business operation
and analytics. I believe I will be able to make contributions to this field of study which I have chosen
by research activities and gathering knowledge through practical works and studies. What draws me
most to this program is courses are fitting best with my interests and experience like -
Computational Transportation, Consumer Research, Machine Learning, Predictive Analytics and
Forecasting and gave me access to many international seminars on topics that will help me to
develop core knowledge for the field.

I have chosen Germany to study because I am familiar with the strong education standards of
German Universities, as my sister-in-law is pursuing a Master’s in Germany and my brother is
working with a German company. The reason why I choose “Otto von Guericke University
Magdeburg” for my studies is because of the curriculum, International diversity in students, and
what attracts me more is the Research and transfer section on the university webpage how students
and faculties are involved in several different research and innovation programs. I am particularly
interested in operations research, business analytics, and entrepreneurship and always wanted to
study in an environment with diverse mindsets around me to exchange ideas, interact and
collaborative learning with students and faculties, bring new thoughts to the table and work on
projects to contribute in the area of study.

I am confident I will learn and gain good knowledge from this course which helps me to achieve
my long term career goals in the field of operations and analytics, also I would be able to
contribute to giving back smart solutions to our community using my skills and learnings from this
course and I am excited to join this program.

B) Summary of your technical skills

Please specify the nature of your Bachelor’s thesis/scientific project:
 Analytical/Numerical
 Empirical/Statistical
 Other, please specify:

Please specify the applied method and key results of your Bachelor’s thesis/scientific
project. (1400 characters
 )

Introduction : The project aims to optimize banking sales services using machine learning and deep
learning models. Three focus areas were used to optimize the bank´s services.

 First is to predict the estimated salary of bank customers and offer them better services
corresponding to their income.
 The second area is to predict customers who might leave the bank and focus more on them so
that they will keep the bank´s services.
 The third area of research is to find a group of similar customers and classify them as
valuable/non-valuable and loyal/non-loyal so that the bank can focus on valuable and loyal
customers.
To achieve this research result, different regression models such as Decision Tree
Regression, Random Forest Regression, Polynomial and Support Vector Regression were used
to predict the estimated salary of the customers. Several classification models such as
Logistic, K-NN, Naive Bayes, Decision Trees Classifier, Random Forest Classifier, and SVM
were used to predict the churn rate i.e. customers who might leave the bank. Later XGBoost
model was used for better performance and faster execution to predict churn rate. A deep
learning model - Artificial Neural Network has been used for better accuracy. To find a
similar group of customers, unsupervised machine learning techniques such as K-Means and
Hierarchical clustering were used. Several different evaluation techniques such as confusion
matrix, R-Squared, Adjusted R-Squared and SSE were used to assess the performance of the
models. K Cross-validation was used to confirm whether a model is overfitting or
underfitting.
Goal : The primary motivation of the project is to implement an efficient algorithm to increase the sales
of the banking business in the future. Without anticipating the actual demands of the customers may
lead to failure of the business. The algorithm targets critical points like customer retention, attrition rate
and other behaviours of the customers. There are many failures in the banking business of many
companies all over the world as there is no proper usage of modelling techniques to predict the
reasons for the churning of customers. Here we need to see whether the client is satisfied with the
services or not otherwise there is a chance of customer to jump to the other company without previous
warning. The leading causes that influence clients are socio-demographic variables, products
contracted, engagement variables, service usage, technical incidents, stationary variables,
and competitor variables. Therefore, there is a need for developing new methods to manage
customers.
The objective of the project is as follows:
To implement deep learning models and XGBoost model to predict the customer attrition rate.
To implement machine learning models like Multiple regression model, Support vector regression,
decision tree regression, random forest regression to predict the customer salary and give offer and
services according to their income.
To implement models like logistic regression, random forest classifier, SVM, decision tree classifier,
naive-Bayes classifier to classify customers churning rate.
To implement models like hierarchal and k-means clustering to find a similar group of customers and
to find a group of valuable/loyal customers.

Research Methodology: Regression analysis is carried out to predict the salaries of customers
so that bankers could provide more benefits or better services according to the incomes of
those customers who are likely to leave their organization. We considered several different
models to predict the salary and check the efficiency of each regression model on our data.
R squared (𝑅2) error which is also known as the coefficient of determination is used as an
evaluation metric for the models. In the regression model, we are trying to minimize the
Research Methodology: Regression analysis is carried out to predict the salaries of customers
so that bankers could provide more benefits or better services according to the incomes of
those customers who are likely to leave their organization. We considered several different
models to predict the salary and check the efficiency of each regression model on our data.
R squared (𝑅2) error which is also known as the coefficient of determination is used as an
evaluation metric for the models. In the regression model, we are trying to minimize the
errors by finding the best fit line.
The architecture of the defined project is composed of five steps. They are: defining the problem, data pre-
processing, selection of model, splitting the dataset into
training and test set, building regression, classification models, clustering and artificial neural
network.
Implementation : Implementation is done in python using several python packages and machine learning
algorithms.

Steps included in our implementation process are as follows:

 Data preparation: The dataset used for our analysis is taken from Kaggle in CSV format. It is a churn
modelling dataset of online banking based out in Europe. The dataset has 10000 observations with 14
attributes (dependent and independent variables) .
 Data Visualization: Dataset has been visualized to get insights into the data.
 Data pre-processing: Data was prepared by ruling out the missing values for attributes. As a second
step correlation was carried out on all the attributes. After careful examination, we found that the first
three independent attributes are irrelevant to consider for predicting the churn (our target was exited
rate). Therefore these three attributes were ruled out and retaining the rest of them for further
implementation. This step is the most important to get good output from the implemented models.
 Concepts implemented: The following concepts were implemented for complete analysis and
understanding of customer churn and customer behaviour. Prediction by using various regression
models, namely: linear regression model, polynomial regression model, multiple regression model,
decision tree, random forest, XGBoost and support vector model.
 Building the models: The processed dataset was used to build and validate the model. The dataset
was split into two subsets: a training set and a test set in the ratio of around 1:3. To build the different
models on the training set, we have used the Scikit-Learn library which consists of several algorithms
and supports NumPy and SciPy library, and for prediction of churn, the test set was used to obtain the
results of correct and wrong predictions. For clustering, it was implemented by scipy cluster hierarchy
and skikit-learn packages of python library. For implementing the artificial neural network, Keras
machine learning library of python has been used.

Conclusion & Discussion : In regression, after applying every possible model for testing the accuracy of the
estimated salary of the customers, we concluded that random forest is working more efficiently when
compared to other models on this dataset. But further improvement is required in data pre-processing (a better
dataset) before fitting the model for getting better results for the R squared score.
In classification, several models were implemented and compared to classify the customers into two classes
(churner or non-churner) and also to predict whether the customer will stop using the
services of the bank and switch to some other bank. The usage of models helps us to predict
churning of customers correctly and identifying the attributes which lead to retaining them. Finally, we can
conclude that the XGBoost classifier model would be the best algorithm for the churn prediction model.
In Deep Learning Model, an artificial neural network is performing well for 1000 epochs but not as much as
the XGBoost model is performing. A higher number of epochs could help model for a better result but the
execution time will grow up exponentially. Clustering was carried out to cluster a similar group of customers
based on different features such as age, balance, credit score, etc. The main aim of clustering was to cluster a
similar group of customers and to investigate most, average and least valuable customers. By doing this,
banks can identify different classes of customers and provide benefits and services to increase customer
satisfaction and to retain customers who may probably churn.