B8101 ~ B7101

Business Analytics 2

Note: unlike the class offered in the Spring, this Business

Analytics 2 class is a full semester class. If you already took
BA2 in the Spring but are still interested in taking this class
to experience the additional material, please contact the
instructor.

Need help? Have any questions? Email

ba2@[Link] – this email address will reach the
RoboTA for the class who will forward your email to me
and every TA, and remind us every hour if we don’t
respond to your email promptly

Note that there is pre-work required before the first

class – details are below, and on Canvas.

Professor Daniel Guetta

Business analytics refers to the ways in which enterprises such as businesses, non-profits, and
governments use data to gain insights and make better decisions. Business analytics is applied
in operations, marketing, finance, and strategic planning among other functions. The ability to
use data effectively to drive rapid, precise, and profitable decisions has been a critical strategic
advantage for companies as diverse as Walmart, Google, Capital One, and Disney. In addition,
many current and recent startups are based on the application of analytics to large databases.
With the increasing availability of broad and deep sources of information – so-called “Big Data”
– business analytics are becoming an even more critical capability for enterprises of all types
and all sizes.

Real estate developers are unlikely to be skilled builders; but unless they understand how
construction techniques work, they are unlikely to be able to do their job well. Similarly, modern
executives are unlikely to need to code models themselves, but unless the understand how
these techniques work, they are unlikely to be able to leverage them effectively.

In this class, you will extend the material you learned in your core Business Analytics class and
apply these methods to new cases in a broad range of industries. In particular, you will
• Extend and deepen your study of the methods you learned in Business Analytics. You
will learn how to use these methods in more unstructured and diverse situations, on

–1–
 complex real-life datasets, and on a broader range of structured and unstructured data
(such as text and image data).
• Learn more complex, powerful, and flexible methodologies for predictive analytics than
those you covered in Business Analytics, such as random forests.
• Introduce a framework that will help you translate business needs and priorities into
analytic problems.
• Learn the language you will need to communicate with data scientists and other
engineers implementing these models.

Much as Business Analytics does, this course emphasizes that the discipline is not theoretical;
we will apply these new methodologies in a number of cases, and use them to develop
increasingly powerful insights and predictive capabilities. Many of the techniques we will be
covering are now considered standard in industry, and developing a good understanding of
them will deepen your ability to identify opportunities in which business analytics can be used to
improve performance, drive value, and support important decisions. For those of you who will
work closely with data science and product teams, the deep knowledge we will develop in this
class will prove invaluable.

This course will not require any coding, and will not require any prior knowledge other than your
core Business Analytics and Statistics classes.

Pre-work
Before class begins, you will be required to install an add-in for the class, prepare for our first
case, and complete a short survey. Details will be posted on Canvas. Anyone who has not
completed the pre-work at least three days before class begins will be removed from the class.

Detailed class plan

Due to the advanced nature of the material covered in this class, we will focus on quality over
quantity, with a strong focus on making sure you understand the concepts in depth before we
move on. We may therefore not cover all the content below.

Cases are listed in red italics font below.

• Module 1: Introduction
o Introduction to the class – what this class is about, and not about
o Using the XLKitLearn add-in
o The three pillars of business analytics – predict, optimize, and explain
o Understanding predictive analytics
• Module 2: Ideas into Analytics
o Going from a business problem to an analytics use case
o The five ingredients of an analytics problem: the target variable, unit of analysis,
features, evaluation metric, and the baseline
o Incorporating time into analytics problems
o Evaluation metrics
o Picking the correct evaluation metric

–2–
 o The Fixed Income Trading case
o The New York City Public Schools case
Module 3: Overfitting Revisited; the Magic of K-Fold Cross-Validation
o A review of overfitting
o The bias-variance tradeoff
o The “Cubic Model” of COVID-19
o Overfitting and K-NN
o Diagnosing overfitting
o Overfitting in the New York City Public Schools case
o K-Fold cross-validation
o Assessing model stability
• Module 4: Regression on Steroids: The Lasso
o The bias-variance tradeoff in linear regression
o Traditional techniques
o Come Rain or Shine: A Simple Weather Prediction Model
o Shrinkage estimators
o From shrinkage estimators to the Lasso
o Variable selection with the Lasso
o Predicting spending behavior with the Lasso
• Module 5: The Case of Cambridge Analytica
o The Cambridge Analytica Case
o Storing BIG data: representing sparse datasets
o Understanding the predictive analytic task
o From likes to traits: How Cambridge Analytica Profiled the World
• Module 6: An Introduction to Decision Trees
o Autonomous Vehicles: The Analytics Behind the Hype
o Linear regression: Important Shortcomings
o The magic of non-parametric models
o An introduction to decision trees
o Trees and the bias-variance tradeoff
o Fitting decision trees: the CART algorithm
• Module 7: Boosted Trees (time permitting)
o Ensembles: hacking the Bias-Variance Tradeoff
o An introduction to boosting
o The Predictive Medicine Case
• Module 8: Random Forests
o The Lending Club case
o Finding the right target variable
o An introduction to bagging
o Revisiting the marketing spend case
o Random forests
o Model interpretability and variable importance
• Module 9: The USPS Case
o The USPS Case
o Dealing with multi-class classification
o An introduction to neural networks and deep learning
• Module 10: Deep Learning and Neural Networks
o Introduction to Deep learning
o Training neural networks

–3–
 o The BinIt case
o Transfer learning
o Smarter gradients
o Loss functions
o Hardware
o More advanced gradient descent
o Architecture: convolutional neural networks
o The [Link] case
o Architecture: LSTMs
o Architecture: Generative models
o Self-supervised learning
o Interpreting neural networks
• Module 11: Text Analytics
o The case of Evisort
o Representing text data: standardize, tokenize, encode
o One-hot encoding
o Embeddings
o Word2Vec
o Transformers and BERT
o The GPT family
o A demo of Latent Dirichlet Allocation
• Module 12: Data Visualization in Tableau
o An introduction to data visualization
o Visualization best practices
o The NYC Garbage Collection Case
o An introduction to Tableau
o The Citibike Case

Time permitting, we may cover one or more of the additional cases below to further illustrate the
concepts in the class
• Appendix A: Analyzing Racial Disparities in Vehicular Police Stops in Philladelphia
• Appendix B: Analytics at the NFL – Predicting Blitzes
• Appendix C: The Mathematics of Exponential Growth: Understanding COVID modelling,
the SIR Model, and the R-0
• Appendix D: Advanced optimization: robust, automated class and exam scheduling
systems
• Appendix F: Understanding the Bootstrap and p-values in the NYC DOE Case

Requirements and Grading

Before class begins, you will be required to complete some pre-work – see Canvas for details.

The class itself will be graded as follows. Please see Canvas for the due dates for each of
these components, and for due dates for the pre-class work required for each module.

• Final exam (40%): the final exam will be multiple choice. It will not require the use of
Excel, or of a computer. A practice final will be provided on Canvas.

–4–
 • Homeworks (35%): Each homework will be based on a real-world application of the
techniques in this class, and will require you to use the tools we will be learning in the
class.

There will be a menu of homework assignments (which will change periodically) and
you’ll have to pick five or six of those assignments to complete, depending on the
iteration of the class you take; details will be given in the first class and on Canvas.

Data science is difficult, and I would be doing a disservice if I made the homeworks
easy. As such, be warned – these homeworks are designed to be difficult. To make
things fair, I will not grade these homeworks based on correctness – instead, I will grade
them based on effort, understanding, and execution on a scale of 1 to 6 using the
following rubric:
o 0 points: no significant effort
o 2 points: some questions tackled; evidence some analysis was carried out on
the data, but perhaps not correctly
o 4 points: all questions tackled; evidence some analysis was carried out on the
data, but perhaps not correctly.
o 6 points: all questions tackled (but perhaps not correctly) and submitted in a
clear, well-presented, and easy-to-follow report clearly explaining the logic
behind the steps you took.
o 8 points (extra credit): outstanding work, not only answering the questions in
the homework and meeting the requirements for 6 points, but also carrying out
further investigations based on the data given. Homeworks completed correctly
in Python would merit this grade.
Note that each of these rubric descriptions require excellence in modelling and
exposition/presentation.

• Attendance and participation (25%): your attendance and participation score will be
calculated as follows
o Punctuality (25%): this part of your grade will be calculated by finding the
fraction of classes you arrive at exactly on time, or for which your absence is
excused.
o Nameplate (25%): this part of your grade will be calculated by finding the fraction
of classes your attend with your nameplate clearly visible, or for which your
absence is excused.
o Attendance (25%): this part of your grade will be calculated by finding the
fraction of classes you attend (even late) or for which your absence is excused.
o Contributions in class (25%): this part of your grade will be calculated based on
my impressions, an on your participation in ad-hoc assignments such as pre-
class work.
Please note that I am very generous with excusing absences – for any reason –
provided you let me know at ba2@[Link] at least an hour before class.

Course Materials
There is no required textbook for the class. There will be cases and slides that will be posted on
Canvas.

–5–
For those of you looking for additional reading, I have found the following three resources to be
excellent:
• Data Science for Business, by Foster Provost and Tom Fawcett. This book is pitched at
the MBA level, and covers many of the topics in this class. It is excellent, but does not go
into quite as much depth as we will.
• The Elements of Statistical Learning: Data Mining, Inference, and Prediction, by Trevor
Hastie, Robert Tibshirani, and Jerome Friedman. This is the bible of machine learning,
written by some of the greatest innovators in the field over the last 20 years or so. It is,
however, very mathematical, and therefore will be out of reach to most MBAs. That said,
if you have a particularly quantitative background and want to dive in much greatest
depth into any of the topics in this class, this is the class to go.
• Business Data Science, by Matt Taddy. This book is also pitched a more advanced
level, and requires some knowledge of statistics, probability, and calculus. For those with
that background, it covers many (but not all) of the topics we will be discussing in our
class, and includes excellent examples.
• Deep Learning with Python, by François Chollet. Also pitched at a more advanced leve,
this book does a great job at really teaching the fundamentals of data science by
showing you how to fit these models in Python. It also does a great job of avoiding
complex mathematics, but it does require a level of coding and numerical sophistication.

Software
This course will require the use of Excel – we will provide you with an add-in called XLKitLearn
([Link]), which will extend the functionality of Excel to cover the topics in this
follow-up elective. You will be asked to install this add-in as part of the pre-work for this class.

Even though this course only requires you to use Excel, the add-in itself will be powered by
Python code. Python has quickly become the lingua franca of business analytics, and those
hoping to enter analytics-related industries will likely carry out further study to deepen their
knowledge of this programming language. Every run of the add-in will output the equivalent
Python code you would need to run to get the same result, so you can implement these
methods in Python if you like.

Absolutely no Python or coding is required to complete this class.

The BA2 Community

I maintain a Business Analytics 2 mailing list for all alumni of the class. When you complete, the
class, you will automatically be added to this list, which I use around 1-3 times a year to foster
community among alumni of the class, update you on the latest and greatest changes to
XLKitLearn, and announce one-off lectures I will be hosting for alumni on topical analytics-
related subjects. (You are, of course, welcome to unsubscribe at any time, though I'll question
your life choices…)

Inclusion, Accommodations, and Support for Students

–6–
At Columbia Business School, we believe that diversity strengthens any community or business
model and brings it greater success. Columbia Business School is committed to providing all
students with the equal opportunity to thrive in the classroom by providing a learning, living, and
working environment free from discrimination, harassment, and bias on the basis of gender,
sexual orientation, race, ethnicity, socioeconomic status, or ability.

Columbia Business School will make reasonable accommodations for persons with documented
disabilities. Students are encouraged to contact the Columbia University’s Office of Disability
Services for information about registration. Students seeking accommodation in the classroom
may obtain information on the services offered by Columbia University’s Office of Disability
Services online at [Link]/docs/services/ods/[Link] or by contacting
(212) 854-2388.

Columbia Business School is committed to maintaining a safe environment for students, staff
and faculty. Because of this commitment and because of federal and state regulations, we must
advise you that if you tell any of your instructors about sexual harassment or gender-based
misconduct involving a member of the campus community, your instructor is required to report
this information to a Title IX Coordinator. They will treat this information as private, but will need
to follow up with you and possibly look into the matter. Counseling and Psychological Services,
the Office of the University Chaplain, and the Ombuds Office for Gender-Based Misconduct are
confidential resources available for students, staff and faculty. “Gender-based misconduct”
includes sexual assault, stalking, sexual harassment, dating violence, domestic violence, sexual
exploitation, and gender-based harassment. For more information, see
[Link]

–7–