Hey, Colin…interesting article!

Can you elaborate on this section:

“ If the dataset in the churn prediction model could be rearranged into a panel format, where we have multiple observations across time for each customer, not only could the leakage issue be resolved, but a much wider array of potential features and models could be considered.”

Both in terms of how it solves leakage and, more generally, how panel data is better suited to churned modeling.