An alternative technique: Simulated Annealing
Simulated Annealing (SA) is a meta-heuristic search method, as Genetic Algorithms (GA).
However, it is faster but it explores less the search space than its concurrent.
It is implemented in QuantMiner for sake of comparison; since SA is fast, it can help conducting preliminary experiments with a large number of attributes.
SA works on many rule instantiations at the same time. It modifies the interval bounds w.r.t. a decreasing amplitude with iterations. At each iteration, if the quality of the rule (the same criterion as for GA) increases, it replaces the old rule by the new one.
SA own parameters are as follows:
- the number of attempts to improve the rule,
- the number of association rules to test simultaneously, that is the number of instances on the same rule pattern that are processed independently but simultaneously.
Computation time depends on these parameters, as well as the quality of the results.
Note that the default values are large enough to improve quality.