My data scientist interview experience at Visa
Though we are living in a super digital era where Chat GPT can give us an answer to almost any other question we get, it cannot explain one’s experience on going through a company’s interview process for sure! So, in this article I am going to talk about my interview process for the data scientist position at Visa.
The process was divided among four rounds, two online rounds on a teams call and the other two onsite (Round 2 & 3). I am guessing each round has some kind of score which helps the interviewer take the final decision on whether to hire a candidate or not.
Round 1
Gather insights from the card transactions dataset.
The dataset had features like the time of transaction, pincode of the place where the transaction was made, customer ID, sector/company for which the transaction was made. Three sections were there in this test, each for Python, SQL and ML modelling ideas, each given 15 mins time to solve. Python & SQL questions were mostly about finding insights from the dataset. They were not very hard if you are a regular coder in Python & SQL.
Round 2
This round was about testing ML theoretical/technical knowledge. One should have a deep understanding on the models that they have worked on to score full in this round. The interviewer will only ask questions on the topics which are related to your resume. In my case it was XGBoost algorithm along with some data pre processing techniques that I had used in my projects. So following were the questions which were asked to me in this round. The interviewer was such a delight to talk to!
How did you finalise the feature set?
What other algorithms did you explore before selecting XGBoost? — I had mistakenly said Logistic regression, (It was a white lie which got caught :P since training a logistic regression model on 500+ features is not a good idea!)
Is there a higher chance of overfitting in boosting algorithms compared to other classification algorithms? If yes, why?
Explain learning rate in XGBoost algorithm. (Most important questions — asked in almost every interview)
What is a box and whisker plot?
Which hypothesis testing to perform to find out if there’s a sampling bias in the dataset or not?
Round 3
In this round the interviewer was the hiring manager, where she gave me an idea on how the role is going to be like, a little about the team as well. She gave me few examples on what kind of data science projects Visa is working on currently. The next part of this interview was all about asking questions based on my work experience & one case study.
How to find out which variables are important in predicting the dependent variable?
Case study: You are working as a data scientist for a telecom company. How will you find out which customers are going to buy an international roaming pack in next three months?
Round 4
This was the last round, an online interview scheduled with the VP of Global Data Solutions team at Visa. I thought it will be a behavioural round but it was not! He asked me questions to gain a high level understanding of my work experience and domain. Then his questions went like:
What is reinforcement learning? Have I used it in any of my projects?
Scenario based questions: Metrics of success of an airline reward program.
How to find out if the control/test group, while performing an experiment on a population, is representing the population fairly?
All the interviewers were professional and made me feel comfortable throughout the interviews. I had prepared for case studies related to banking sector but none of them were asked. Though the data science case studies are very similar in the sense of finding metrics to measure success, the feature set used to solve them can differ based on their domain.
All in all it was a good experience. Let me know if you have any more questions!
