What is Data Exploration?

Data exploration provides a first glance analysis of available data sources. Rather than trying to deliver precise insights such as those that result from data analytics, data exploration focuses on identifying key trends and significant variables. 

Data exploration is also referred to as exploratory data analysis. 

How is Data Exploration Used in Enterprise?

Data exploration is often used as a first step before more resource-intensive analytics efforts. This approach is useful when working with large, unfamiliar data sets, or when the analytics team is trying to figure out where to look for useful insights. 

Data exploration helps to answer two important questions about data:

  • What are the most important variables?
  • What are the relationships between those variables?

An example of how this might be useful in business is if a company discovers that there is a recurring variation in average customer spend. This value might correlate with another value, such as a special price offer at the start of each month. By identifying this correlation, analysts know where to focus when looking for insights into customer behavior. 

Data Exploration Methodology

Data exploration is generally automated, especially when working with big data. Exploration can be done using tools such as Microsoft SandDance or MIT’s open-source DIVE.

In some instances, analytics engineers may use manual techniques to perform exploratory data analysis on smaller data sets. This is done with data exploration tools, by coding applications in Python or R, or simply by viewing the data in an application like Excel.

Whether manual or automated, data exploration uses several statistical techniques to identify significant variables, such as:  

Univariate Analysis

Univariates are variables that can provide insight without requiring further context. These variables fall into three categories:

  • Discrete: Numerical values, such as the number of customers per day. Analysts may look at the sum, average, or count of unique values. 
  • Continuous: Numbers on a relative scale, such as percentages. Analysts will often focus on the range of these values (the difference between highest and lowest.)
  • Categorical: Recurring text values, such as the state in a U.S. customer’s address. Analysts often count the frequency of these variables. 

Correlation Analysis

Many variables have some degree of correlation with another, and this can be measured with statistical techniques, such as comparing the standard deviations of each variable. 

Rather than investigate every correlation, analysts focus on the ones with the highest rate of correlation. Often, they create a grid or matrix to provide a unified view of correlations, allowing them to identify the most promising relationships.

Cluster Analysis

Clustering helps to identify a correlation that applies to some data, but not all. In manual exploration, this can be done visually by plotting the data as a scatter chart. Possible correlations will appear as dense clusters in the resulting chart – hence the name. 

Clusters can also be the basis for segmentation, which is when the analyst focuses on a discrete subsection of the data for the purposes of further exploration. 

Data exploration can sometimes yield actionable insights of immediate value to the business. In most cases, however, this type of analysis serves as a jumping-off point for more intensive analytics projects.

Glossary of Terms

A guide to the nomenclature of data integration technology.