Project paper, code, and presentation slides: https://github.com/justmarkham/kaggle-allstate
Goal : Predict which car insurance options a customer will buy.
Problem:
(1) There are 7 car insurance options, each with 2 to 4 possible values.
(2) Values are identified by number (0,1,2, etc)
(3) “quote” consists of a single combination of those 7 options.
(4) Customers review one or more qutotes before making their purchase.
dataset:
quotes history and purchase history.
diffculty:
there are 2304 possible combinations of options.
no partial credit. no feedback given on which options were wrong.
naive approach:
for each customer, just predict that they will purchase the last set of options they were qutoed.
Accuracy vs random guessing.