Apriori advantages and disadvantages

Advantages

• Simple and easy to understand and implement.
• Uses the Apriori property (all subsets of a frequent itemset are frequent) to prune the search space heavily, so it is far better than testing all 2ⁿ itemsets.
• Easy to parallelise (candidate counting is independent) and works well with sparse data and a high minimum support.
• Produces results that are complete (all frequent itemsets are found).
• Works on large transaction databases, and the many variations (hashing, partitioning, sampling) can improve it.

Disadvantages

• Multiple database scans: one for each itemset size k, which is costly for large databases.
• Large number of candidates: for 10⁴ frequent 1-itemsets, more than 10⁷ candidate 2-itemsets are generated, and very long patterns need exponential candidates.
• High computation and memory for candidate generation, and for subset testing against every transaction.
• Becomes very slow with low minimum support or dense data with long frequent patterns.
• Needs min_sup and min_conf to be set by the user; wrong values give too few or too many rules.
• Treats all items equally (no weights or quantity), and many generated rules may be uninteresting or redundant.

FP-growth overcomes the candidate generation and repeated scanning problems.