Akeed Restaurant Recommendation Challenge
Predict which restaurants in Oman customers will most likely order from
Prize
$3 000 USD
Time
Ended over 2 years ago
Participants
242 active · 1348 enrolled
Helping
Oman
Prediction
Customer service
Dataset and challenge objective misalignment
Data · 24 May 2020, 07:23 · 5

In the description of the challenge, "the objective ... is to build a recommendation engine to predict what restaurants customers are most likely to order from given ... the customer order history." However, we are given no indication of the order history of customers in the test set. In fact, there are zero customers in common between the training set and the test set. How have you worked around this?

Discussion 5 answers

You are correct. The workaround is to treat this like a cold-visitor recommendation. Try to cluster customers based on features they have in common, then recommend the most popular restaurants in train for the customers in test in the same cluster.

Another way is to treat it strictly as a classification problem.

I totally agree with you.

The provided training data has lots of features that can't be used as they aren't available as part of the testing data.

Exactly! The features available to us won't produce good predictions as many users are missing gender (and even those that do, there's a very tiny sample of female users) and dates of birth.

Even, i feel the same..there are not many features available..Also, it doesn't look like a recommendation problem...It looks like a classification problem...