A Five-Step Guide for Conducting Exploratory Data Analysis

Summary

Exploratory data analysis (EDA) is presented as a practical way to describe a dataset, uncover patterns such as missing data, correlations, and outliers, and generate hypotheses that inform business questions and later data products. The guide recommends first clarifying the intended task with leadership and product, then checking dataset size, sample rows, and the unique row identifier; its synthetic merchant-behaviour example has 1,500 samples, 13 features, and one row per shop per day. It then advises examining missingness by feature, investigating why values are absent before deleting or imputing them, and classifying features as continuous, discrete, or categorical to guide analysis. Visualizing distributions and relationships helps reveal changes across samples and time, while box plots and percentile thresholds support systematic outlier review. The process is framed as a starting point whose findings can support regression, classification, or KPI-dashboard decisions.

Context

The guide addresses the challenge of describing an unfamiliar dataset and using EDA to inform business questions, data products, dashboards, and possible future modeling tasks. It emphasizes gaining context about the intended outcome before deciding what to examine.

Approach / What changed

The proposed framework begins with dataset context, size, sample rows, and unique identifiers, then examines missing values, classifies features, visualizes distributions and relationships, and investigates outliers. The examples use a synthetic merchant-behaviour dataset and include feature-specific handling such as imputation or a “Not Answered” category.

Takeaways

• High missingness should not automatically lead to removing a feature: understanding why values are absent can reveal meaningful nonresponse and prevent bias. Handling methods depend on the data, including interpolation for trending time series and a separate category for unanswered survey responses.
• Classifying features as continuous, discrete, or categorical determines which visualizations and statistical methods are appropriate. The example distinguishes GMV, AOV, and conversion rate from count-based measures and finite-value fields.
• Box plots can reveal outliers, while comparing the 1st and 99th percentiles provides a systematic approach for continuous features. Percentile-based detection does not apply to discrete features, which require different interpretations or methods.

Related reading