writeup.txt

After looking through the example data the best criteria
found were that during the 4-noon time period there were
very few robo calls. Next to none actually. So assuming
calls during that time are not robocalls is relatively
accurate. Also if the number called more than about 5 people
it is most likely a robo caller. There are of course exceptions
to this but it is mostly true. Instead of finding criteria to
deal with these exceptions I assumed if they were making calls
to more than five numbers then it definitely was a robot. 

It became more difficult to differentiate robo and human when
a number only called one person one time. This did not happen
very often for robo callers but it did happen. Because I could
not determine a viable solution to this I assumed if the number
only called one number then it was a human. 

By filtering on the 4-noon calls I would manage to
scoop up the humans who called lots of numbers and not include
them as robots.

The code is written in Python and requires the pandas module. pandas
makes working with large data sets simple. The algorithm is unfortunately
slow, but the goal of this was accuracy not speed so I did not worry about it.
Speed is lost when setting the LIKELY ROBOCALL column.