IE Masters in Analytics and Big Data – first hand report

First hand report on Master in business analytics and big data program at IE (Madrid, Spain) - why, what, how, days, and challenges.



By Nicky Sarof.

Why IE?
My journey in data science began with an undergraduate course in data mining. It opened me up to a world of possibilities. The brilliance of it was that it could be used to solve a wide variety of problems such as identifying fraud, grouping products that were often bought together etc. and thus, better understand human behaviour and the only catch was that there needs to be data and patterns to find. When the opportunity to work at an analytics firm presented itself, I jumped at it!

After working for a few years, I decided that I needed to broaden my horizons and after much deliberation, I chose to attend the IE School master in business analytics and big data program at IE. I chose the program because of:
  • Breadth of topics being covered - providing 360 Degree view of Analytics and its relevance in business
  • Support for entrepreneurship
  • The school’s reputation
  • Strong alumni network
  • Diversity of the student body

Course Curriculum
The curriculum itself is balanced and comprehensive that covers the current data science trends and an in depth coverage of business (telecom and utilities, finance, marketing etc.) and technical components including big data technologies such as spark, HDFS etc. There is a lot of emphasis on working in teams, development of business communication skills with a special focus on problem solving using real world cases.

Day at IE
A typical day at IE includes 5-6 hours of classes, group work and individual work. The classes are conducted by eminent professors who are leaders and experts in their respective fields, and bring with them a wealth of experience in dealing with Real World Analytics Problems. As such we not only learn about the different techniques and algorithms, but also how and when those techniques and algorithms are used in solving real world business problems.

Group Work
Groups generally have a broad mix of people from different places and diverse professional backgrounds which helps bring different perspectives and ideas when solving a problem. It is also easily one of the most important components of the program. Individual work involves a lot of self-study to back up the learning from the classes.

Modelling Challenge
As a part of the, we are exposed to a lot of challenging courses. One such course was Financial analytics. As a part of our group project for the class, a contest (with competition leader board) was held wherein 5 different teams competed to come up with the best model for detecting fraudulent transactions.

Some of the key components of the exercise for coming up with the best model were-
  • Right variable selection (from 4 different sets-Ordinal, nominal, continuous and categorical)
  • Variable Transformations
  • Application of right algorithm/technique - Logit, trees etc. with or without ensembles, boosting, random forests etc.
  • Validation on Out of Time Sample using measures such as GINI and KS (Kolmogorov-Smirnov)

Gini Score: Area between the curve and the random model

KS score - maximum difference between cumulative Percentage of bad vs good customers

Different teams used different strategies. While some focused on using all the variables with a standard algorithm, some focused on using the right algorithm/technique in conjunction with right variable selection. On the leader board, it soon became apparent that having the right set of variables is as important as the model building itself.

During the course of Model Development, we realized that different techniques are affected by data in different ways – OLS Regression for instance is highly affected by outliers and non-linearity in the data. We also discovered that the right method of evaluation of the model is as important and why an over fitted model performs badly on out of time sample. By the end of the module, we had developed sufficient know how in tackling and building a model on a real world dataset

Conclusion
The biggest learning for us being this far into the program is that analytics sits on an interesting intersection of different fields and in order to be successful, it requires the ability to not only adapt quickly to change but also apply different problem solving approaches/algorithms to solve business problems. There is no one size fits all and finding the right combination of trade-off between accuracy and other measures is often very challenging.

Nicky Sarof Nicky Sarof is an analytics consultant with a background in computer science and the recipient of the Big Data Asia Scholarship, currently pursuing his Master in Business Analytics and Big data at IE (Madrid, Spain).





Related: