Data Scientist Interview Questions

Responsibilities

Data scientist responsibilities may include:

• Data mining using industry best practices
• Augmenting existing data-collection databases
• Building classifiers using machine learning techniques
• Creating automated tracking systems
• Verifying the integrity of data used for analysis

Skills

Data scientists are essential for gathering data about an organization or industry. In order to obtain useful information and utilize it effectively, a skilled data scientist will:

• Possess a strong work ethic needed to carefully examine large amounts of data
• Possess an eye for detail to catch inaccuracies and outliers
• Communicate clearly with staff members and senior members alike
• Compile information into identifiable documentation
• Possess organizational skills to stay on top of competing priorities

Qualifications

Most positions as a data scientist require applicants to possess a master’s degree in mathematics, computer science, or a related field. However, it is not uncommon for data scientists to possess doctorate degrees. 

Candidates may also obtain additional certification or go through advanced learning courses to further differentiate themselves within the field.

If you’re getting ready to interview for a position as a data scientist, you can prepare by researching the company as much as possible. Learn about the 9 things you should research before an interview.

Salary

Salaries for data scientists range between $87K and $131K with the median being $107K. 

Factors impacting the salary you receive as a data scientist include:

• Degrees (associate's or equivalent technical training, bachelor's, master's)
• Years of Experience
• Location
• Reporting Structure (seniority of the manager you report to and number of direct reports)
• Level of Performance - Exceeding Expectations

Interviews Are Unpredictable

Be ready for anything with the interview simulator.

Get Started

Data Scientist Interview Questions

Question: What steps do you take to ensure the regression model fits the data?

Explanation: This is a technical question. As a data scientist, you can anticipate that the majority of the questions you will be asked during a job interview will be technical. Technical questions should be answered succinctly and directly with no embellishment.

Example: “There are several steps you can take to ensure the regression model fits the data. The first is to employ R-squared methodology. This involves the relative measure of fit. The second is to use the F1 score to evaluate the null hypothesis. The final methodology is RMSE which provides the absolute measure of fit.”

Question: Can you describe what a decision tree is and how it is used?

Explanation: This is another technical question. This question asks you to define a term and provide an example of how it is used in your profession. This is a typical structure of technical questions. Your answer should address the definition first and then provide an example of how you would use this item in your job.

Example: “Decision trees are a graphical model used to illustrate the options available and choices made during a decision process. Like a tree, it begins with a base and expands. Each decision option is known as a node. When you reach the top of the tree, the last decision options are known as leaves. While a decision tree is intuitive and easy to build, it lacks accuracy.”

Question: Do you believe many small decision trees are better than one large one, and if so, why?

Explanation: The interviewer is asking a follow-up question to the previous one. During an interview, you should anticipate follow-up questions. By keeping your answers short and to the point, you enable the interviewer to ask follow-up questions or move on to another topic.

Example: “No, just the opposite. The larger the decision tree, the more accurate it is. Small decision trees lead to problems with fit because the options are few. Ideally, your model would look more like a forest than a tree with many options and a clear path navigating through the forest.”

Question: Why do you think mean square error is a bad measure of model performance?

Explanation: This is yet another technical question. When you answer a technical question, you should anticipate a follow-up question. Follow-up questions indicate the interviewer is interested in the topic they are asking you about. This signals the topic is important to them and that you may want to spend more time on your answers to these questions.

Example: “I do believe that the mean square error or MSE is a bad measure of a model’s performance.  The issue is that the MSE weighs large errors more than small ones.  This puts too much emphasis on large deviations in the data.  A more robust model is the mean absolute deviation or MAE.”

Question: Can you describe some of the assumptions required for linear regression?

Explanation: This technical question is asking for several items as part of your answer. Providing a list of items in an answer is a common practice during an interview. Make sure you organize your answer in a clear manner without repeated items.

Example: “There are several assumptions required for linear regression analysis. These include:

• The data used in the sample is representative of the population

• The relationship between X and the mean of Y is linear

• The variance of the residual is the same for any value of X

• All observations are unique and independent of each other.”

Question: Why is it important to do data wrangling and data cleaning before applying machine learning algorithms?

Explanation: This is an operational question. The interviewer will ask operational questions to learn more about how you do your job. You can answer this type of question by walking the interviewer through the process step by step. Make sure you don’t go into too much detail. The interviewer will ask a follow-up question if they need more information.

Example: “It is important to do data wrangling and data cleaning before applying any machine learning algorithms. This ensures the data sets are appropriate and the actual data sets the analyst intended to work with, the relationships between the data are valid, the standard deviations meet the study guidelines, and the data is standardized and normalized, removing any outliers or variables that would skew the results.”

Question: What are some of the shortcomings of a linear model?

Explanation: The interviewer is asking another technical question but in a back-handed manner. They are asking you to point out a negative aspect of the topic they are addressing. Be sure to stay positive when you answer this question. Going too negative will reflect poorly on you, even though you were asked to discuss shortcomings.

Example: “A linear model has several drawbacks. First, it holds some strong assumptions that may not be true for the application being used. It also assumes a linear relationship, normality between the variables, minimal multicollinearity, and homoscedasticity. In addition, a linear model cannot be used for discrete or binary outcomes.”

Question: What steps do you take to deal with an unbalanced binary classification?

Explanation: This is yet another operational question asking how you react to a specific situation that may occur during a data analysis exercise. As an experienced data scientist, you should be able to answer this question easily.

Example: “The most obvious way to deal with unbalanced binary classification is to consider the metrics you are using in your model.  Some metrics will skew the results even though they are accurate. Another way is to increase the penalty for incorrectly classified and any minority-class data. This will result in a better model with more accurate findings. Finally, you can oversample some of the minority-class data or undersample some of the majority-class data, thereby balancing the classification.”

Question: Can you describe the differences between a box plot and a histogram?

Explanation: The interviewer is asking another question of a technical nature. This one is asking you to compare different types of visual models used to analyze data. As a reminder, technical questions are best answered by comparing the terms presented by the interviewer and then possibly providing an example of how they are used in your profession. Technical answers should be brief and to the point.

Example: “Boxplots and histograms are similar in that they are visualizations used to illustrate the distribution of the data. However, they communicate information in different ways. Histograms are bar charts that illustrate the frequency of a numerical variable’s values. This enables the viewer to understand the shape of the distribution, the variation, and any potential outliers. Boxplots don’t allow you to see the shape of the distribution, but you can view other information like the quartiles, the range, and outliers. Boxplots are better than histograms when you are comparing multiple charts.”

Question: What is cross-validation, and how do you use it when analyzing a data set?

Explanation: This is another technical question that asks for both the definition of the term and an explanation of how it is used. During an interview, you want to make sure to listen carefully to the questions you are being asked. Many candidates will start thinking about the answer as soon as the interviewer begins to ask the question which will cause them to miss some critical points and not provide the correct answer.

Example: “Cross-validation is used to assess how well a model performs on a new and independent dataset. A common use of cross-validation is splitting the data into two sets - one to build the model and one to test it.”

Additional Data Scientist Interview Questions

• While compiling a report for user content uploads, you notice a spike in September. What do you think may have caused this?

• Can you explain what data leakage is as it pertains to machine learning?

• How can you test that a feedback survey was filled out randomly or truthfully by customers?

• How do you calculate variance in an unsupervised model?

• Why is ensuring that data is secured so important?

• Name one way data could change the world.