Page 12 - Playtest_Square_Enix - EphemeralData

Feature Engineering

Objectives :
T his process is the most important in this analysis since we have in

the original dataset only descriptive variables, and they also
suffer from a scaling effect that make them not comparable.

That's why our goal will be to create new variables and features that
must help us put a weight on the original features of the dataset.

For example, is a fail from the first checkpoints the same as a fail from

latest checkpoints in the game ? Are users more likely to replay, quit or
fail at the beginning or at the end ? To answer these questions we need
new features that help weight users' action through the entire game.

Creation of new features:

Replayability Tenacity Perseverance

Total pass
Total fail
Total quit
________________ ________________ ________________

Unique u Unique u Unique user fail
ser pass
ser quit

Replayability : After passing a checkpoint, how many times will I repass
again ? The higher this feature the higher is the level of replayability of
a checkpoint.

Tenacity: Even though I quitted once, how many times I tried and
quitted again but did not give up. The higher this feature the higher is

the level of tenacity of users.

Perseverance: I already failed once, but how many times I tried and
failed again but did not give up. The higher this feature the higher is
the level of perseverance of users.

All these features are very good indicators of the dedication of a user

considering a checkpoint. And what's very interesting is that they are
scale down on unique user which mean they can be compared between
checkpoints.
11