Explore jobs
Find specific jobs
Explore careers
Explore professions
Best companies
Explore companies
Data science is a highly technical field that requires a lot of knowledge and expertise. Asking the right interview questions for a data scientist can seem daunting, but there are myriad resources to help select the right ones.
Key Takeaways:
As a candidate, it’s very important to go over the necessary knowledge, terminology, and area of expertise before going into an interview. A lot of the interview questions will be about making sure that the candidate is qualified for the position.
Qualifications are very important to determine but don’t forget to ask other types of questions. While data scientists work alone a lot, they do share their findings with other members of the team, so they need strong interpersonal skills.
There are a number of common data scientist interview questions. With a highly technical profession like this, many of them will be technical questions as a way to ascertain the candidate’s qualifications, but that doesn't mean that data scientists won’t be asked behavioral interview questions or general questions. Here are some example answers.
Tell me about yourself
This is an opening interview question to try to assess the candidate's personality and cultural fit. It’s also a way to tell how interested the candidate is in the profession and how long they’re likely to stay at this job. As a candidate, it’s best to emphasize your passion for the field as well as your deep interest in it.
Example Answer:
Well, I’ve been working with data for three years now, and so far, I’ve found it very fulfilling. I fell in love with statistics in high school and decided to get a degree in it. Now I get to work with stats all day, as well as teach other people about it. Overall, I’d say I’m rather pleased with my career trajectory.
In terms of my hobbies, I love making models. There’s nothing quite like the satisfaction of finishing one and getting it just right, in my opinion. I also love trivia. I go to trivia events with my friends regularly.
What methods do you use to identify outliers in a data set?
In statistics, outliers are values that don’t fit in with the majority of the data. Outliers can skew statistical analysis if they’re far enough away from the median, meaning that it’s important that data scientists be able to identify them and know how to categorize them.
Example Answer:
That depends a lot on the data I’m working with. The first thing I do is analyze the data. What model and method I use will depend on the size of the data set, the nature of the data, and what I’m trying to pull from it. From there, I’m able to select the model that allows me to determine which data points are outliers.
How do you use a confusion matrix to calculate accuracy?
Confusion matrices are a way to categorize data. When predicting values, you can end up with more than two categories, which means that a classification algorithm won’t be helpful.
Example Answer:
For most data, we’re working with a prediction. So, that means I’m trying to find out if the result we ended up with matched what the prediction was. That means that you can have a true positive, true negative, false positive, or false negative.
If I put all these into a confusion matrix, you can see how many times the data didn’t end up the way you predicted. And you can see which way it went – if you got a negative when you expected a positive, and vice versa. But the formula for calculating accuracy is: Accuracy = TP + TN / TP + TN + FP + FN
What do you know about random forests?
A technical question like this is a way to identify if the candidate has the knowledge and skills to do the job. Being able to explain and possibly implement a random forest is a way for candidates to show that they have expertise on the subject.
Example Answer:
Random forests are learning algorithms. They’re a way to generate more accurate predictions. When I make a random forest, I start by building a bunch of decision trees based on samples of the data I’m working with.
Then I add splits to it, choosing sample predictors. Decisions should be based on the majority rule in order to get the best results.
Explain cross-validation
There are myriad data science techniques in order to analyze data and determine accuracy. Interviewers want to be sure that candidates are familiar with these different techniques and able to explain them.
Example Answer:
Cross-validation is a way to test the accuracy of a predictive model. It’s a resampling method, so you take one data set as a learning set for the algorithm, then create a test set for it to make predictions with. Then you see how accurate the predictions are to the data.
The idea is to try to find problems like selection bias or overfitting. But all of these prediction models are only as good as their algorithms. And the data, of course. You have to make sure to give it good data to start with, or you’ll get bogus predictions. After all, garbage in, garbage out, right?
Why do you want to work here?
While the majority of people go to work for a paycheck, it’s better if the candidate is actually interested in working at the organization that’s interviewing them.
Changing jobs is disruptive for all parties, so it’s best for workers to prioritize applying to places where they actually want to work and for employers to hire people who actually want to work there.
Example Answer:
I’ve been working in the private sector, and while I admit I like the paychecks, it’s time for a change. Not only am I at a point where I want more stability, but I like the idea of my work making a difference to people. Not that what I was doing didn’t exactly, but analyzing how to sell things just didn’t really make me feel like I was improving society.
What are the steps involved in maintaining deployed models?
If the organization that’s hiring makes use of deployed models, then it would behoove the hiring manager to ascertain whether or not the candidate is familiar with them. It’d be best if they’d used them before, but so long as they’re familiar with the practice, they ought to be able to learn how to use them quickly.
Example Answer:
Monitor, evaluate, compare, and rebuild. The first allows you to determine how accurate the model is, as well as the impact of processes. The second is to make sure that you don’t need to change or update the algorithm – you’re getting the results you want, and they make sense. Third, you compare it to other models to find out which one is most effective.
Lastly, you take your best-performing model and rebuild it for the current data sets. I usually end up doing a bit more tweaking here and there to make sure that the model is the best it can be, but that all depends on the data set, how much time I have, and what exactly I want from the algorithm.
What is selection bias, and what are the various types?
Bias is something that can skew the results of statistical analysis, so it’s something that data scientists need to be familiar with and know how to identify and avoid. Interviewers want to be sure that the candidate takes it seriously and knows how to deal with it.
Example Answer:
Selection bias is one of those things that almost everyone has heard of, but most people can’t explain. It makes for some interesting discussions at parties. Selection bias is when the results are skewed because of bias in the selection. For instance, if the sample isn’t selected randomly and researchers are able to select the subjects themselves.
The types of selection bias are:
Sampling bias. Which is when the sample isn’t taken randomly. That will lead to certain members of the group being excluded and/or overrepresented.
Attrition. This is when parts of the study end up being omitted or not completed. This can be because people dropped out of the study, tests weren’t completed, or some trial subjects ended up being discounted.
Time interval. Trials can end at an extreme value, which may end up not being useful.
Data. Often this is the result of cherry-picking. The people running the test have an agenda, so they end up selecting the results that fit with it.
What are your strengths and weaknesses?
Self-awareness is something that can make a candidate appealing. That being said, a candidate that’s a veteran of the interviewing process will take the weakness question and turn it into something that can instead be seen as a strength.
Example Answer:
My strength I’d say is my ability to focus. I can really dial into what I’m doing and get a lot done in a short amount of time. Since most of what I do is solitary, being able to tune everything else out is extremely useful and allows me to stay focused on my task and prevents me from having to keep spinning back up.
As for my weakness, I guess I’d say that I can be a bit forgetful. I get so focused on what I’m doing that sometimes I’ll end up not following up with people. I’m not antisocial or anything – I’ve always gotten along well with coworkers – I just sometimes need reminders to pull me out of the zone.
What do you think are the qualities that a data scientist needs to be successful?
Asking a question like this is a combination of a personality question and a work style question. There isn’t necessarily a “right” answer; however, there are likely certain qualities that the interviewer will be looking for. It’s important to see if the qualities that the candidate prioritizes are similar to the ones the interviewer does.
Example Answer:
I would say that the qualities most successful data scientists have are being excellent with numbers, attention to detail, and good programming skills. Certainly, those are the skills that have served me well in my career.
How do you overcome challenges to your findings?
How exactly the candidate will act in the workplace is impossible to know until they’re hired. However, asking hypothetical questions about how they’d act in certain situations can give you an idea of their personality and whether or not they’ll fit in in the organization’s culture.
Example Answer:
I assume you mean conflicting data. Because if someone just says that they think I’m wrong when I’ve got graphs and charts up to prove my point, then I usually just avoid engaging them. Because clearly, data isn’t the thing that’ll change their mind.
However, if someone comes up with an alternate interpretation of the data or has a valid data set that contradicts my findings, then I usually start by asking questions. I want to know what methods they used, how they checked their data, and all that. I’ll also explain what I did to see if I made an error of some sort.
Sometimes there are just different ways of looking at the data. Or findings from it that weren’t immediately obvious, partially if the point was to look at one particular aspect of the data. And that can be a really interesting and fun discussion to have.
What drew you to a career in data science?
While knowing more about the candidate’s youth or journey to their current career isn’t likely to be a make-or-break question, it can tell the interviewer more about them as a person. Asking this particular question can also be a way to determine how likely they are to stay with their current career path.
Example Answer:
I fell in love with statistics in high school. I took an AP stats class, and I just loved it. I’d never really liked math that much before – I was good at it, but it didn't really speak to me. It was just so removed from everything. It kind of felt pointless. But with statistics, it became really obvious what its uses and applications were.
So I got a degree in it. And I still loved it, even if I didn't always like it. So when I found out that there was a job you could do where you work with statistics all day and get paid a good salary, I was like: sign me up!
What resources do you use to stay up to date on new data science skills?
As technology positions require candidates to stay up to date with the latest advancements and popular techniques, it’s important to find out how they plan to do that. For interviewees, this should be a softball question – so long as they’re actually keeping up with new tech as it comes out.
Example Answer:
I read up on it a lot. I subscribe to the International Journal of Science and Analytics and the Data Science Journal. I also check up with Data Science Central and some other general science magazines and sites, like Science.com. I also attend seminars and conferences when I have the chance.
Why should we hire you?
Interviewers are impressed by interviewees who have done their research on the position they’re interviewing for, so it’s best to try to hit the traits that were mentioned in the job description. It’s also another way to assess personality and find out if the candidate is a good fit for the company culture.
Example Answer:
I’ve been interested in working in the public sector for a while now, though I ended up finding a good opportunity to learn some skills in the private sector. But I’m ready for a shift, and I think that the stability of the position will really help me improve. And I would really like to make a full career out of this job if I get the opportunity.
How do you effectively manage your time?
The nature of the job is largely solitary, so the interview questions for data scientists should reflect that. As a rule, people in this position spend a lot of time working alone without any particular supervision. That means that they have to be able to be self-motivated and able to manage their time well.
Example Answer:
It’s weird: I’ve always been good at time management, but when people ask me how I do it, it’s like… I just do things that need to be done when they need to be done. But what I primarily do is make liberal use of my calendar. I set to know what my deadlines are, and in order to make them, I get an idea of what needs to be done and when.
So I guess you’d say that I set attainable goals for myself in order to reach the bigger goal in the end.
What do you do when you get something wrong?
How an employee will behave at work is something that is difficult to determine by a resume or even during an interview. That being said, asking candidates behavioral questions can give interviewers an indication of how the candidate will behave and whether or not they will fit in with the current culture.
Example Answer:
Well, when I mess up, I try to figure out why I messed up. That’s very important to me so that I don’t make that mistake again. Now, exactly what else I do will depend on the nature of the mistake. If my screw-up caused problems for someone else, I’d be sure to apologize. If someone has to get in trouble over it, I’ll fess up, no question.
As a rule, though, I just try to correct my mistakes and keep going. I’ve always had a tendency to beat myself up about them, and I’m trying to break it. Everyone messes up. I’m way more forgiving when other people make mistakes than I do unless it’s just carelessness, of course.
For text analytics, would you prefer Python or R?
In order to be a successful data scientist, the candidate will need to be familiar with some programming languages. Python and R are both regularly used, so it’s good to check to make sure that the interviewee is familiar with them and familiar with their strengths and weaknesses.
Example Answer:
I’m better with Python, so I tend to gravitate towards that language. It’s really great when working with large volumes of data. However, in unstructured data, R can just be better. I’m working on improving with it, as it definitely has times when it is just the better option.
What are your hobbies outside of work?
In most cases, what an employee does outside of work shouldn’t affect their chances of getting hired. But if the employer is looking for a cultural fit or just wants the candidate to open up, asking about hobbies is a good way to get them to start talking. People are generally willing to have a conversation about the things they enjoy doing.
Example Answer:
I love building models. I’m into model cars right now, but I’m thinking of getting into ships. I’ve done all sorts, though; I've got quite the collection in my basement. I’m also a trivia buff. I love learning new things and factoids, and I go to trivia night with my friends regularly.
What is the purpose of A/B testing?
There are several different types of tests that can be performed. Data scientists don’t just analyze data; they’ll also gather it. A good candidate needs to be familiar with different types of tests, how to conduct them, and their purposes.
Example Answer:
A/B testing is primarily used in marketing. So, the purpose is to determine which is more effective at getting the response you want: A or B? In order to find this out, most tests will swap out one thing on a website, poster, or sales email and see how the subjects react to it. Do they respond better to A or B?
Where do you see yourself in five years?
The candidate’s goals should be reasonable and in alignment with the employers’. The majority of interviewers want to know that the interviewee has looked ahead to the future and has some ambition.
Example Answer:
I really like data analysis, so I’d like to keep doing it. I’m hoping to be a senior analyst in the next five years, but I’m not quite sure where to go from there. I think I’d be content to keep working with data. So I’d like to see myself in a senior position in the next five years.
What is root cause analysis?
Root cause analysis is an important aspect of data analysis. It focuses on what the actual cause of the problem is, rather than just the observable symptoms. Having a data scientist that is willing to dig deeper into the data to look for the cause is a good thing, as well as being sure that they’re familiar with the concept.
Example Answer:
Root cause analysis is a type of problem-solving. It’s often easy to see all the smaller problems and fix those without realizing that they all stem from a single, bigger cause: the root cause. Root cause analysis describes the methods used in order to find that root cause.
Here are some additional questions to ask in a data scientist interview.
What is the purpose of data cleaning in data analysis?
What are linear regression and logistic regression?
What is deep learning?
Why is TensorFlow considered important in data science?
What is an epoch?
Explain overfitting and underfitting
What is a computational graph?
What are vanishing gradients?
What is Ensemble learning?
How do you treat outliers?
How often would you update our algorithms?
What’s your preferred programming language?
What’s the law of large numbers?
How would you describe recommender systems?
Have you ever declared a time series as stationary?
As a Candidate:
Be sure to double-check all the definitions of relevant terms, such as data cleaning, A/B testing, random forests, and confusion matrices. Brush up on your programming knowledge as well because the interviewer may want a work sample in the interview to prove your credentials.
Go over your cover letter, resume, and the job description. Make sure you're prepared for questions directly about your resume or cover letter. Also, be ready to reference the qualities and qualifications listed in the job description when answering questions.
Check out common questions that data scientists are asked, such as the ones in this article. While you don’t want to completely script your answers, it’s good to be prepared for the common types of questions you’re likely to be asked.
As an Interviewer:
Make sure that you’re familiar with the position’s requirements and needed expertise. If you don’t feel qualified, then make sure that you have an expert in the room with you to verify the answers the candidate gives are correct.
Review the candidate’s resume and cover letter, as well as the job description. Be prepared to ask specific questions about them if they seem applicable.
Prepare the questions that you want to ask the candidate. It’s often better to over-prepare in terms of having too many questions, but don’t try to stuff them all into the interview if you find out you run out of time.