Hello there,
I am currently working on a machine learning project and could use some guidance on feature engineering. The project involves predicting customer churn in a subscription-based service.
It includes features such as customer demographics, subscription plan details, usage history, and customer service interactions.
The target variable is a binary indicator of whether a customer churned or not.
I have already done some basic preprocessing, such as handling missing values and encoding categorical variables. However, I’m struggling with creating meaningful features that could improve the model’s performance.
I have also gone through this: https://community.drivendata.org/t/feature-engineering/3396sap which definitely helped me out a lot.
Which feature engineering techniques have been effective for similar churn prediction tasks?
Are there any specific interactions or transformations of features that are known to be useful?
How should I handle time-dependent features, such as customer tenure or recency of interactions?
Also, can someone provide any insights or resources that could help me improve my feature engineering process and ultimately enhance the predictive performance of my model.
Thankyou in advance.