Gym Membership Dataset: A Complete Data Dictionary Audit

By shrijeetverma13 · July 18, 2026

This analysis examines the structure and documentation quality of a 24-column gym membership dataset, revealing that integer columns dominate (75%) and…

The 24 columns in the dataset fall into three data types: integers, strings, and floats. Integer columns dominate with 18 columns (75%), followed by string columns with 4 (16.7%), and float columns with just 2 (8.3%).

The columns with the most detailed descriptions—and therefore likely the most business-critical—are: churned by february (the target variable tracking cancellations), primary goal (customer objective categories), joined with friend (referral/accountability flag), distance km (proximity to gym), and goal aggressiveness (goal intensity rating). These fields stand out because their descriptions are longer and more explanatory, suggesting they play a key role in understanding customer behavior and predicting outcomes like churn.

The columns with the shortest, least detailed descriptions are 'sex' (13 characters: "F / M / Other"), 'member id' (16 characters), 'age' (19 characters), and both 'week4 visits' and 'week1 visits' (20 characters each). These brief descriptions are strong candidates for expanded documentation to help users better understand what each column represents.

The most common data type in this schema is 'int' (integer), used in 18 out of 24 columns, making up 75% of all columns. This strongly suggests the dataset is quantitative and measurement-focused, meaning it's well-suited for statistical analysis, calculations, and numeric modeling rather than text-based categorization.

Description length varies by column data type: float columns have the longest average descriptions at 38.5 characters, followed by integer columns at 32.5 characters, and string columns have the shortest at 29.25 characters.

The analysis explored how column names are structured across different data types. It generated three visualizations: one showing the number of columns for each data type, another comparing the average column name length by type, and a third highlighting the top 10 most common suffixes (the last word in each column name) used across the dataset. These charts together reveal how naming conventions like word count, length, and suffix choice vary by data type.

The analysis searched the 'description' column for keywords like 'TBD', 'unknown', 'n/a', and empty strings to flag potential data quality risks. Based on the results, no columns were flagged as having unclear or missing documentation, meaning all descriptions currently in the dataset appear complete and well-documented.

After calculating the word count for each column's description and applying the IQR (Interquartile Range) method, no columns were found to be statistical outliers. This means all the descriptions have a fairly consistent length, with none standing out as unusually short or long compared to the rest.

The analysis generated a comprehensive overview of your data dictionary, including visual breakdowns of column type distribution and key summary metrics. Two charts were created: one showing the distribution of column types across your dataset, and another summarizing key metrics like total columns, unique types, average description length, and the ratio of ID/Date-related columns versus other columns.

The analysis grouped your dataset's columns by their 'type' field and organized them into structured tables showing each column name alongside its description. It also examined the descriptions to identify which ones reference categorical value sets (patterns like 'F / M / Other' that use '/' characters to separate possible values). Two data tables were generated as part of this analysis, giving you a clear structured view of columns by type and their categorical value indicators.

The sentiment analysis of the description text found that most entries (83.3%) are neutral in tone, with only a small fraction showing positive (4.2%) or negative (12.5%) sentiment. The average polarity score is nearly zero (-0.003), indicating an overall balanced, neutral tone across the dataset.

A word cloud was successfully generated from the 'description' column, visually displaying the most frequently occurring words in your data. Words that appear larger and more prominent in the cloud indicate higher frequency in the dataset, making it easy to spot dominant themes and common terms at a glance.