If your organization is transitioning to an AI-first strategy, you already understand the importance of data. The effectiveness of your algorithms and machine learning models depends on the quantity and quality of available data.
But raw data rarely arrives neatly organized or ready for analysis. Hidden patterns, outliers, missing values, and inconsistencies can all influence the insights you uncover and decisions you make. That’s why exploratory data analysis (EDA) is a critical step in any machine learning workflow.
The guide below explains what exploratory data analysis is, why it matters in machine learning, and the key steps involved in the process.
Featured Certificate Course
Data Science and AI for Decision Making
Learn how leaders can use AI and data science to improve decision-making, innovation, and business performance.
Start Anytime | 4 Modules | 5-6 Hours per Module | $1,949
What Is Exploratory Data Analysis?
Exploratory data analysis (EDA) is the process of examining a dataset to identify patterns, distributions, and quality issues before applying algorithms or machine learning models.
“Exploratory data analysis is how we begin to understand the shape, structure, and quirks of the data—before making any assumptions about what models to use or what patterns might emerge,” says Harvard Business School Professor Iavor Bojinov in the online course Data Science and AI for Decision Making.
According to Bojinov, EDA helps organizations address several key questions before applying algorithms to a dataset, including:
What variables are present in the data?
Are the variables categorical, numerical, or mixed?
How are the variables distributed within the dataset?
Do the variables differ enough to be useful in machine learning?
Does the dataset contain outliers, missing values, measurement errors, inconsistencies, or unexpected patterns?
Is the dataset truly representative, or are there sampling biases?
EDA goes beyond data wrangling or cleaning. It’s a broader investigative process that helps you understand the insights a dataset may reveal during machine learning.
“Good EDA is part technical, part investigative,” Bojinov says in Data Science and AI for Decision Making. “You’re not just cleaning the data—you’re practicing modeling hygiene: getting to know the data, looking for structure, spotting strange behavior. And always circling back to your business question: What might this tell me?”
Related: Listen to Professor Bojinov discuss AI adoption, trust, and decision-making on The Parlor Room podcast, or watch the episode on YouTube.
Why EDA Is Essential to Machine Learning
Without exploratory data analysis, machine learning models can produce results that are technically accurate but ultimately ineffective.
“Once you’ve defined the problem and gathered your data, it can be tempting to jump straight into modeling,” Bojinov says in Data Science and AI for Decision Making. “But skipping [exploratory data analysis] is one of the fastest ways to end up with a model that’s technically correct—and practically useless.”
EDA helps uncover issues that can affect model performance before analysis begins. You might identify missing or inconsistent data, detect sampling bias, or discover patterns that reshape how you approach the problem altogether.
It can also surface unexpected insights. An outlier, for example, may signal a data error—or reveal an important subgroup worth further investigation. EDA helps determine the difference before those issues influence your results.
Exploratory Data Analysis Steps
1. Evaluate the Data’s Quality
Before beginning exploratory analysis, confirm that your dataset is reliable enough to support meaningful insights.
Start by identifying missing or incomplete values. If values are missing across the dataset, it could indicate a loading or data collection issue. If they appear only in specific entries, they may point to unique subgroups, edge cases, or inconsistencies.
Addressing these issues early can improve analysis accuracy and help prevent misleading results later.
2. Check for Outliers
Next, check the dataset for outliers. In Data Science and AI for Decision Making, these are defined as “extreme values that differ significantly from the typical range” within a dataset.
Outliers can signal errors in your data, but can also reveal meaningful trends or subgroups. For example, if your organization provides a digital platform, unusually high usage patterns may indicate a segment of heavy users who engage with the platform differently than the average customer. Understanding those behaviors could surface valuable business insights.
To determine whether outliers should be removed, retained, or analyzed separately, you need to perform an investigation. During that process, consider which variables the outliers affect and how far they deviate from expected patterns in the dataset.
3. Define and Analyze Variables
Next, determine which variables to include in your analysis. This decision should be guided by:
The business problem or question you’re trying to solve
The variables available in the dataset
Your professional understanding of which variables are important
After defining your variables, you can begin analyzing them. Here, you’ll want to measure the typical range of values and start identifying trends, relationships, and anomalies.
Analysis can be univariate, meaning you investigate a single variable to spot patterns and insights. Or, it can be multivariate, meaning you analyze two or more variables together to understand how they interact. Both can aid the machine learning process by helping you determine which models to deploy.
4. Visualize the Data
Much of the exploratory data analysis process involves working with raw numerical data. While numbers can reveal important insights, they can also obscure patterns—especially in large datasets.
That’s why Bojinov recommends making data visualization a standard component of any EDA effort.
“Visualization is where it often starts,” Bojinov says in Data Science and AI for Decision Making. “Not just to plot, but to see. To ask: What’s really going on here?”
To illustrate the visualization's importance, Bojinov points to a famous example: Anscombe’s Quartet. Created in the 1970s, the quartet consists of four datasets that have nearly identical summary statistics. Each has the same mean, variance, and correlation—but when visualized, they take on dramatically different shapes.
.png)
“Numbers alone can hide what’s really happening,” Bojinov says in Data Science and AI for Decision Making. “Visualization grounds us in reality. Ground the analysis in the patterns you see. Because if you don’t know what’s in your data, your model won’t either.”
Free Course Flowchart
Which HBS Online Digital Transformation & AI Program Is Right for You?
This flowchart will help you determine which HBS Online digital transformation and AI course best aligns with your career level and experience, learning objectives, and educational goals.
One Step in the Machine Learning Life Cycle
Exploratory data analysis is critical for ensuring that your machine learning efforts extract the most value from your data, but it’s ultimately one step of a larger process.
After evaluating data quality and outliers and understanding your variables, you still need to transform the data into something usable for machine learning—and, finally, apply an algorithm to model the data.
If you’re new to using AI and data science to make decisions and inform your organization’s strategy, learning frameworks and processes like exploratory data analysis is essential. To jumpstart your efforts, consider completing a course like Data Science and AI for Decision Making.
Are you ready to improve your ability to leverage data and artificial intelligence in your business? Explore Data Science and AI for Decision Making—one of our online digital transformation and AI courses—and download our free flowchart that helps you choose the right course for your goals and needs.