Develop gym customer churn prediction models.
- Perform in-depth analysis of customer churn data.
- Determine the influence of correlation data.
- Train a model to cluster clients into precise classification groups.
- Create a profile of customers with a high probability of churn.
- Train a model for customer churn prediction.
- Achieve a recall of 93% on test data.
The final model for predicting customer churn is based on Gradient Boosting. The 'recall' metric of the test data is 0.937.
The client who is more likely to leave the fitness center is likely to have the following characteristics:
- Has 2 months or less remaining in their contract.
- Does not participate in group visits.
- Is under the age of 26.
- Has a low level of average additional charges.
- Has a lifetime of only 2 months since their first visit.
- Has a low average weekly class frequency (both in general and in the current month).
- Data preprocessing
- In-depth data analysis
- Machine learning model development
- Predictive modeling
- Clustering modeling
- Python: pandas, NumPy, Matplotlib, Seaborn, Scikit-learn, SciPy (for hierarchical clustering)
- Clustering techniques (Dendogram, Elbow method, Silhouette, KMeans, Agglomerative Clustering, DBSCAN)
- Prediction Churn Model
- Clustering Clients Profiles Model