As pointed out in the journal paper about EFIM
published in KAIS:
- EFIM is hard to beat in terms of memory usage for high utility itemset mining
- It is also hard to beat EFIM for dense datasets
- But for sparse datasets, EFIM is not always the fastest.
So, a limitation could be that it is not always the fastest on sparse datasets.
In data mining a lot of researchers focus on the speed of algorithms. But performance is not everything. Actually, features offered by an algorithm are probably more important for users than speed. There are a lot of additional features that could be added to EFIM.
- For example, it would be possible to modify EFIM to use length constraints (as in the FHM+ paper, where length constraints allows to further reduce the search space using some new properties).
- It would be possible to make a version of EFIM for items with negative utility, an incremental version of EFIM, a version of EFIM for on-shelf high utility itemset mining etc.
Actually, many ideas could be combined with EFIM to make some new algorithms with more features.
Edited 2 time(s). Last edit at 03/16/2017 01:42AM by webmasterphilfv.