Gym Membership Churn Analysis

Live Dashboard

Code on GitHub (coming soon)

Project overview

This project analyzes 4,000 gym members to answer three questions: when are members most likely to cancel, what behavioral and account factors predict it, and can a simple, explainable rule flag at-risk members before they leave.

Source: Gym Customers Features and Churn (Kaggle)

Size: 4,000 members, 14 original features (demographics, contract details, attendance behavior, spend, and churn outcome)

Data quality: Audited for nulls, duplicates, dtype errors, and logical inconsistencies (e.g., months remaining exceeding contract length). Zero issues found. No cleaning required, only verification.

Methodology

Explore

Analyzed membership tenure, visit frequency, and engagement patterns to identify behaviors that precede cancellation.

Model

Built a churn-risk scoring approach in Python based on tenure and behavior signals: plain logic, validated against outcomes, no ML required.

Visualize

Surfaced at-risk segments in a two-page Tableau dashboard (an executive overview and a segment deep-dive) so an operator can see who's likely to churn and why.

Findings

The first thing that jumped out: this is a first-90-day problem, not a loyalty problem. 42% of members who cancel do it within their first 3 months, and the rate drops to near zero after month six.
Contract length matters on its own, separate from tenure. 1-month contracts churn at 42%, 6-month at 12%, and 12-month at just 2%.
One habit roughly cuts risk in half: group classes. Members who attend churn at 17% vs. 33% for those who don't.
Proximity mattered more than I expected. Members not located near the gym churn at 40% vs. 24% for those who are.
I didn't need a model to make this actionable. A 3-factor rule got me most of the way there. Flagging anyone at ≤3 months tenure, on a 1-month contract, not attending group classes, catches a group nearly 5x more likely to churn than everyone else. Plain logic, no ML required.

Recommendations

Fix onboarding before worrying about loyalty. Build a structured 90-day check-in program, since risk basically disappears after that window on its own.
Get new 1-month members into a group class in week one. That single nudge roughly halves their churn odds.
Use the at-risk flag as an actual retention worklist, not a metric that sits in a report. 1,013 members are identifiable today with data that's already on hand.

What this demonstrates

Turning a churn problem into a simple, explainable rule rather than reaching for ML by default.
Tableau proficiency rounding out the stack (Power BI covers PP-02/PP-03).
Recommendations tied directly to specific, actionable member segments.