#1 Alternative testing of predictions

open
nobody
Recommender (2)
5
2009-02-26
2009-02-26
Jaap Reitsma
No

Currently the recommender framework uses time-based testing. An alternative is to use 80/20 testing: Use 80% of the dataset (random selection) to learn from, then use the remaining 20% to validate the prediction performance. This could be even repeated with new 80% samplings for cross-validation.

Discussion