Prepare for and practice interview questions from Brandeis University.

Brandeis University Interview Questions

Brandeis University Interview Guides

Machine Learning

How would you encode a categorical variable with thousands of distinct values

Encoding Categorical Features

Interpret coefficients of logistic regression for categorical and boolean variables

Coefficients of Logistic Regression

How does random forest generate the forest and why use it over logistic regression

Random Forest Explanation

Replaced by QUESTION 791 - Should be deleted

Bias Variance Tradeoff

Explain the difference between XGBoost and random forest and give an example where you would use one over the other

Xgboost vs Random Forest

Behavioral

Describing a data project and its challenges

Hurdles In Data Projects

Strategically resolving misaligned expectations with stakeholders for a successful project outcome

Stakeholder Communication

How would you answer when an Interviewer asks why you applied to their company?

Why Do You Want to Work With Us

Making data-driven insights actionable for those without technical expertise

Simple Explanations

What do you tell an interviewer when they ask you what your strengths and weaknesses are?

Your Strengths and Weaknesses

Statistics

Explain what a p-value is to someone who is not technical

P-value to a Layman

What are the assumptions of linear regression?

Assumptions of Linear Regression

How would you tackle multicollinearity in multiple linear regression

Multicollinearity in Regression

Determine whether using an 83% cutoff overestimates or underestimates the actual population’s credit score

Credit Score Estimation

Data Structures & Algorithms

Write a function to return the optimal friend that should host the party.

Optimal Host

Swap the values of `a` and `b` in the dictionary without declaring any other variable

Swap Variables

Write a function to find and return the last node of a singly linked list. If the list is empty, return null.

Last Element of a Singly Linked List

Analytics

How would you design and measure the impact of a job training program on employability?

Job Training Program Evaluation

Describing a real-world data cleaning and organization project

Data Cleaning Experiences

Given a table called <code>user_experiences</code>, write a query to determine the percentage of users that held the title of <code>&#34;Data Analyst&#34;</code> immediately before holding the title <code>&#34;Data Scientist&#34;</code>.

Immediate is defined as the user holding no other titles between the <code>&#34;Data Analyst&#34;</code> and <code>&#34;Data Scientist&#34;</code> roles.

Example:

Input:

<code>user_experiences</code> table

<table>
<thead>
<tr>
<th>Column</th>
<th>Type</th>
</tr>
</thead>

<tbody>
<tr>
<td><code>id</code></td>
<td>INTEGER</td>
</tr>

<tr>
<td><code>position_name</code></td>
<td>VARCHAR</td>
</tr>

<tr>
<td><code>start_date</code></td>
<td>DATETIME</td>
</tr>

<tr>
<td><code>end_date</code></td>
<td>DATETIME</td>
</tr>

<tr>
<td><code>user_id</code></td>
<td>INTEGER</td>
</tr>
</tbody>
</table>
Output:

<table>
<thead>
<tr>
<th>Column</th>
<th>Type</th>
</tr>
</thead>

<tbody>
<tr>
<td><code>percentage</code></td>
<td>FLOAT</td>
</tr>
</tbody>
</table>

User Experience Percentage

Tell me about a data project that didn’t go the way you expected. What did you set out to do, what surprised you, and how did you handle it?

How would you explain what a p-value is to someone who is not technical?

What does a p-value in a statistical test represent?

What does a p-value in a statistical test represent?

P-value to a Layman I

In a statistical test, how does a low p-value (less than 0.05) influence our decision about the null hypothesis?

In a statistical test, how does a low p-value (less than 0.05) influence our decision about the null hypothesis?

P-value to a Layman II

Let’s say you have a categorical variable with thousands of distinct values, how would you encode it?

Which method can be used to extract communities from large networks, that does not require a pre-determined number of clusters like K-means?

Which method can be used to extract communities from large networks, that does not require a pre-determined number of clusters like K-means?

What are the assumptions of linear regression?

Which assumption of the residuals of the standard linear regression model can not be overcome by increasing the sample size?

Regression assumptions

Let’s say we’re trying to determine fake reviews on our products. 

Based on past data, 98% reviews are legitimate and 2% are fake. If a review is fake, there is 95% chance that the machine learning algorithm identifies it as fake. If a review is legitimate, there is a 90% chance that the machine learning algorithm identifies it as legitimate.

What is the percentage chance the review is actually fake when the algorithm detects it as fake?

What is the percentage chance the review is actually fake when the algorithm detects it as fake

Fake Algorithm Reviews

Probability

How does random forest generate the forest? Additionally, why would we use it over other algorithms such as logistic regression?

What happens when you average the output on multiple decision trees?

What happens when you average the output on multiple decision trees?

Average Trees

How would you interpret coefficients of logistic regression for categorical and boolean variables?

Why is one-hot encoding recommended for categorical variables in logistic regression models?

Why is one-hot encoding recommended for categorical variables in logistic regression models?

Let’s say you’re working with a city agency that’s rolling out a new job training program for unemployed adults. The program includes skills workshops, career counseling, and job placement assistance, and you’ve been asked to evaluate whether it actually improves participants’ employment outcomes over the next year.

How would you design a study to measure the program’s impact on employability, and what metrics and methods would you use to ensure your evaluation is both rigorous and fair?

You’ve been tasked with building a classification machine-learning model to predict whether a transaction is either fraud or not fraud for a credit card company. You have ten years of historical data on transactions, including a flag for whether a transaction was manually identified as fraud.

Describe how you might go about building a fraud detection model for credit card transactions. Be sure to mention the possible model types, discuss the bias-variance tradeoff in model development, and address any complexities that arise from the class imbalance.

Input features in the data include:

<ul>
<li>Transaction amount</li>
<li>Merchant category for the transaction</li>
<li>Zip code for the merchant</li>
<li>Zip code for the billing address</li>
<li>Average transaction amount for the account over the past six months</li>
</ul>

Output feature:

<ul>
<li>0/1 flag for fraud (0 = not fraud, 1 = fraud)</li>
</ul>

Note: Fraudulent transactions are (thankfully) a very small percentage of all historical transactions. Assume fraudulent transactions are 0.01% of historical data.While building the model to perform the classification, you need to consider the bias/variance tradeoff, and take into account the fact that there is a class imbalance (very few of the observations are “fraud”).

Bias variance tradeoff and class imbalance in finance

Bias - Variance Tradeoff and Class Imbalance in Finance

Let’s say you’re a data engineer at Fidelity Investments, and you’re running a SQL query on a cloud-based data warehouse. All cluster resources and network health metrics look normal, but the query is still taking over 10 minutes to complete.

How would you go about diagnosing and improving the performance of this query?

How would you diagnose and speed up a slow SQL query when system metrics look healthy?

Slow SQL Query

Query Optimization

You are given a dictionary with two keys <code>a</code> and <code>b</code> that hold integers as their values.

Without declaring any other variable, swap the value of <code>a</code> with the value of <code>b</code> and vice versa.

Note: Return the dictionary after editing it.

Example:

Input:

<pre tabindex="0" class="chroma"><code>numbers = {
 &#39;a&#39;:3,
 &#39;b&#39;:4
}
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>def swap_values(numbers) -&gt; {&#39;a&#39;:4,&#39;b&#39;:3}
</code></pre>

How would you handle the data preparation for building a machine learning model using imbalanced data?

Addressing imbalanced data in machine learning through carefully prepared techniques.

Data Preparation for Imbalanced Data

You work as a backend engineer at Glassdoor. The company relies on multiple legacy systems to aggregate and process job postings, salary data, and company reviews. To ensure reliable data ingestion and processing, you are tasked with building a monitoring service that tracks the health of these legacy systems via regular heartbeat signals.

The service must expose one endpoint to receive heartbeat signals from each legacy system, and another endpoint to return a list of systems that have not sent a heartbeat in the last five seconds. How would you architect this monitoring system? Write a technical design document describing your approach.

Design a monitoring service for legacy system health via heartbeat endpoints at Glassdoor.

Legacy System Heartbeat Monitor

DevOps

How would you tackle multicollinearity in multiple linear regression?

You’re building a machine learning model to predict hospital readmission risk using patient records from the past five years. The data includes patient demographics, diagnoses, treatment history, and outcomes, but hospitals periodically change treatment protocols and data collection standards over time.

How would you structure your training, validation, and test splits to ensure your model generalizes to future patients? What risks would arise if these datasets were not clearly separated, and how would you validate that your final evaluation reflects real-world performance?

How would you structure your training, validation, and test splits to ensure your model generalizes to future patients? 

Training vs Validation vs Test Data

Let’s say we’ve deployed a recommendation model for an e-commerce site, and after two months, we notice that the model’s click-through rate has dropped from 12% to 7%.

How would you investigate the cause of this performance decline, and what steps might you take to address it?

How would you investigate and address a decline in a model's production performance?

Model Product Performance Degradation

ML Ops

As a data engineer for Slack, they asked you to design their new product, “Slack for School”. When designing their database, you ponder upon the following questions:

<ol>
<li>What are the critical entities, and how would they interact?</li>

<li>Imagine we want to provide insights to teachers about students’ class participation. How should we design an ETL process to extract data about when and how often each student interacts with the app?</li>

<li>Suppose a teacher wants to see the students’ assignment submission trends over the last six months. How would you write a SQL query to retrieve this data?</li>
</ol>

How would you answer?

System design for a digital classroom service.

Digital Classroom System Design

Data Modeling

When an interviewer asks a question along the lines of:

<ul>
<li>What would your current manager say about you? What constructive criticisms might he give?</li>
<li>What are your three biggest strengths and weaknesses you have identified in yourself?</li>
</ul>

How would you respond?

When asked about your strengths in an interview, what is an effective way to respond?

When asked about your strengths in an interview, what is an effective way to respond?

Your Strengths and Weaknesses I

Which of the following is an acceptable strategy when discussing weaknesses in an interview?

Which of the following is an acceptable strategy when discussing weaknesses in an interview?

Your Strengths and Weaknesses II

You are given a singly linked list, write a function to find and return the last node of the list. If the list is empty, return null.

When an interviewer asks you a question along the lines of:

<ul>
<li>Why did you apply to our company?</li>
<li>What are you looking for in your next job?</li>
<li>What makes you a good fit for our company?</li>
</ul>

How should you respond?

When asked 'What are you looking for in your next job?' in an interview, how can you tie the company's employee benefits into your response?

When asked 'What are you looking for in your next job?' in an interview, how can you tie the company's employee benefits into your response?

Why Do You Want to Work With Us I

How can company values be used effectively in an interview when asked 'What makes you a good fit for our company?'

How can company values be used effectively in an interview when asked 'What makes you a good fit for our company?'

Why Do You Want to Work With Us II

When responding to the question 'Why did you apply to our company?' during an interview, what aspect should you highlight?

When responding to the question 'Why did you apply to our company?' during an interview, what aspect should you highlight?

Why Do You Want to Work With Us III

A manager in a packet filling line is complaining that their machine is not functioning correctly. The machine weighs a group of packets at once and attempts to get 25 packets into a box.

He has been getting complaints from customers that the boxes contain amounts other than 25. How would you help out?

How would you help ensure each box contains exactly 25 packets

Incorrect Packets

Business Case

Let’s say we have a group of \(N\) friends represented by a list of dictionaries where each value is a friend name and their location on a three dimensional scale of (\(x, y, z\)). The friends want to host a party but want the friend with the optimal location (least distance to travel as a group) to host it.

Write a function <code>pick_host</code> to return the friend that should host the party.

Example:

Input:

<pre tabindex="0" class="chroma"><code>friends = [
 {&#39;name&#39;: &#39;Bob&#39;, location: (5,2,10)},
 {&#39;name&#39;: &#39;David&#39;, location: (2,3,5)},
 {&#39;name&#39;: &#39;Mary&#39;, location: (19,3,4)},
 {&#39;name&#39;: &#39;Skyler&#39;, location: (3,5,1)},
]

def optimal_host(friends) -&gt; &#39;David&#39;
</code></pre>

Describe a time when you had to define a long-term vision for a project or team and move it from concept to reality.

Interviews for leadership or senior technical roles look for more than just “finishing a project.” Your answer should specifically address:

<ol>
<li>The “Why”: What data or organizational gap necessitated this vision?</li>
<li>The Framework: How did you translate a high-level goal into a roadmap?</li>
<li>The Friction: How did you handle stakeholders or team members who were skeptical of the new direction?</li>
</ol>

Vision Setting and Execution Strategy

Product Sense & Metrics

Talk about a time when you had trouble communicating with stakeholders. How were you able to overcome it?

How would you convey insights and the methods you use to a non-technical audience?

Explain the difference between the XGBoost and random forest algorithms and give an example where you would use one over the other.

Why might a data scientist choose to use XGBoost instead of Random Forest for a particular machine learning task?

Why might a data scientist choose to use XGBoost instead of Random Forest for a particular machine learning task?

XGBoost vs Random Forest: Choice

Tell me about a project in which you had to clean and organize a large dataset.

Let’s say that you work at a job board.

A PM wants to build a “related jobs” feature on every individual job description page. This “related jobs” feature would be a sidebar of jobs that are most related to the current job the user is browsing.

You have a couple of ideas on how to find related jobs. There’s NLP concepts such as bag of words and word embeddings. The PM also clarifies that the definition of “related jobs” are other jobs that are similar in position title and job description.

However when presented with the problem, you’re quickly realizing that there are millions of new jobs posted each day on the job board, and finding the top 10 related jobs for every single job posted each day could be extremely inefficient.

Explain a system / method that could solve the problem of finding the top 10 closest related jobs for millions of new jobs per day?

Note: Assume an existing pool of tens of millions of jobs that could be related for each new job. Also assume you have access to all text features of a job such as title, description, company, date, etc..

Explain a system/method to find the top 10 closest related jobs for millions of new jobs per day

Related Jobs Optimization

Let’s say you are given a model that predicts whether a piece of news is relevant or not when shared on Twitter.

How would you evaluate the model?

Calculate Moving Average	SQL	Easy
Predict Customer Churn	Machine Learning	Medium
A/B Test Significance	Statistics	Medium
Optimize Query Performance	SQL	Hard
Feature Importance Analysis	Machine Learning	Medium
Clean Missing Data	Python	Easy
Neural Network Architecture	Deep Learning	Hard
Calculate Cohort Retention	SQL	Medium
Bayesian Probability	Statistics	Easy
Recommend Similar Products	Machine Learning	Hard

Brandeis University Interview Questions

Brandeis University Interview Guides

Brandeis University Interview Questions

Challenge

Brandeis University Salaries by Position

Discussion & Interview Experiences

Discussion & Interview Experiences