Predictive features - unleash your creativity

An old hand at machine learning and data sciences will tell you that she eats, sleeps and drinks feature engineering when it comes to building solid predictive models. We document here some of the interesting features tried out by the kids at the camps. As with any other experimental science, some are successful and some not. The features listed out here should motivate mentors at subsequent camps to get the kids to think along these lines and open their minds to the possibility of such ideas having an impact while building their models!

In the current setting. while introducing features, we motivate how the name (whether it's an old-sounding or a new-sounding name), hobby (whether it's an indoor or an outdoor activity) and gender affect how friends are made. Kids and mentors are then encouraged to go beyond these features and explore others which might signal friendship.

A note - the features listed below may have been motivated by looking at one particular data set which the kid was analyzing. The discrimination percentage mentioned below may not generalize to other data sets.

The information is presented in the following format -
[Feature description] - [Camp] - [Discrimination on the friend-set]*

• Artsy vs Non-artsy hobbies - Bangalore - 88%
• Happy vs Grumpy looking faces - Pune - 55%
• Weird vs Common name - Pune 
• Hobbies involving hand held tools vs otherwise - Pune

*Please read the experiment details to understand this better.