Let's say we work at a credit card company and we're trying to figure out if people are going to pay their bills on time. We have everyone's purchases, split into four main categories: groceries, dining out, utilities, and entertainment. What are some ways you might use KNN to create this model? What aspects of KNN would be useful? Write up your thoughts in submit a link below.


There are a few ways KNN could be applied.


With some exceptions (vacations/trips, or other special occasions), purchases are reasonably predictable--periodic, similar in number, and on average, close to the same amount per period. This should be especially true for utilities, and mostly true for groceries. Outliers here would be numbers of charges above normal levels, and/or charges in amounts above what is typical (measured in distance from neighbors of the same category).


Purchases categorized as entertainment or dining out might be less predictable but should also remain within a reasonable limit over time--this limit would be calculated per customer, representing their budget. Outliers in this category would look similar to those for utilities and groceries, but probably given a little less weight. For example, I might shell out for expensive tickets to a concert for myself and 3 friends. That charge would be well above my usual entertainment budget, but a rare enough occurrence that it shouldn't be of concern unless it becomes a pattern that is above my budget limit.


A third factor would be payment history. Since charges are aggregated monthly, it can be difficult to know exactly when a charge is paid off, unless the customer regularly pays their entire balance every month. If a customer doesn't pay their entire bill every month, maybe charges can be weighted based on a rolling amount of payments made, with oldest charges aging into "payment" first. Outliers here would be charges of the same category or amount suddenly taking longer to pay off than their older, paid-off neighbors.


So, some signs that a customer is going to have trouble paying their bills would be deviations from what we know of their purchasing habits:
* More frequent purchases on credit total, and per category
* Higher individual charge amounts, weighted by category
* Charges that remain "unpaid" for longer than normal