Understand which bookings turn into attendance, and what this could mean for class capacity. Start with observed behavior.
Rates below describe the selected cohort. Group sizes matter.
Compare categories, weekdays or AM/PM. Explore their interaction in the matrix.
Observed proportion at each lead time, with descriptive 95% intervals. No predictive model is fitted.
Descriptive tenure bands for the selected cohort. Unadjusted for class and schedule.
Dataset-wide observations. These findings stay fixed when filters change.
Attendance becomes a utilization measure when it is linked to the actual capacity of each class session. This lab explores the framework; local session data is the next step.
This example is hypothetical. The 82% assumption is unrelated to the GoalZone attendance rate. It does not justify releasing extra places.
Link each booking to a class session. Add class capacity, booking and class timestamps, cancellation events and verified attendance. Measure the gap from booked demand to actual occupancy before evaluating calibrated predictions or waitlist decisions.
The author-referenced CSV is the analytical source. The normalized Parquet matches all 1,500 records on shared fields. The 15-row CSV is a preview; the Kaggle notebook is a reference. Supplied files remain unchanged.
Attendance rate = attended booking records / all booking records in the selected cohort.
Non-attendance = recorded outcome 0. Cancellation and confirmed no-show cannot be separated.
Unavailable: session identifiers, capacity, exact class times, calendar dates, member identifiers, cancellations and waitlists. Actual occupancy and empty places cannot be measured.
Statistical scope: unadjusted associations. Wilson intervals assume independent booking outcomes; repeated-member dependence cannot be checked. n < 30 is a display caution, not a validation threshold.
Product scope: public-data exploration. No Optimo model, causal claim or automated overbooking recommendation.
Explore participation, favorite classes and shared interests across member profiles. Preferences suggest hypotheses for programming.
Favorites are multi-select. Class shares use group participants as their denominator.
Complete, unordered preference sets with two or more classes. Single-class favorites are excluded from this table.
An exact two-class set differs from a pair contained in a three-class set. Pair overlap below counts both.
For each tier, percentage of its group participants naming a class. These are preferences, not bookings.
Number of group participants naming both classes, even when they name a third. Diagonal = members naming that class.
Reported frequency, PT use and habitual check-in provide context for the selected cohort.
Group participants / all members at each reported visit frequency.
PT and group participation are overlapping member attributes.
Share of group participants within each habitual gym check-in band. Check-in is not the scheduled class time.
Four mutually exclusive profiles based on reported group-class and PT use. These are descriptive rules, not fitted clusters.
A → B means “among members naming A, how many also name B?” It is not a behavioral sequence.
Use popular favorites to prioritize demand checks. Compare stated interest with bookings and attendance before adding sessions.
Investigate shared-interest pairs against the actual timetable. Preferences alone cannot establish a collision or an ideal class time.
Explore class and PT offers for relevant members. Measure incremental uptake with an opt-in test and a comparison group.
Dataset A: Which bookings turn into attendance? Dataset B: Which members report participating, and which classes share stated interest? Combine the concepts with local member, booking and session data; do not combine the public datasets as one population.
Favorite labels are preserved exactly as supplied, including HIT, LesMiles and Kickboxen. They are not automatically mapped to GoalZone categories. Source files remain unchanged.
Participation is reported: attend_group_lesson is a member-level flag, not a dated attendance log.
Preferences are stated interest: no observed scheduling conflict, class attendance frequency or cross-sell outcome is available.
Source verified: the author describes simulated membership data for EDA, licensed CC0. These generated profiles do not describe Optimo members or validate real-world behavior.
Privacy in this view: dates of birth, gender, supplied age, trainer names and source IDs are excluded from the dashboard payload and cohort export.
Next data: stable member IDs linked to chronological visits and class bookings, session schedules, attendance, capacity, cancellations and timestamped preferences.
Whole-dataset results, n = 1,500. These estimates stay fixed when cohort filters change.
Standardization gives every category the same mix of newer and longer-standing bookings. It is a descriptive adjustment, not a forecast.
The estimates describe this sample. They do not establish causality or validate a model for Optimo. The confidence range shows resampling uncertainty under the stated assumptions; it is not a guaranteed range for another gym.
Whole-dataset synthetic sample, n = 1,000. These results stay fixed when cohort filters change.
Synthetic data demonstrates analytical technique. The pattern estimates and their confidence ranges belong to this artificial sample. Real programming decisions require local member and class records.