Combining indicators - backtesting to find rules

Hello!
There are many (probably) working strategies based on few indicators. For example:
IF (MACD > ...) AND (price grows) AND (5 EMA > 34 EMA) AND (...)
THEN the price will (probably/usually) move up

Some examples eg. here:
[Dear Guest/Member you can't see link before click here to register]
[Dear Guest/Member you can't see link before click here to register]

My idea is to:
make a historical database of: date, time, price, indicator values, etc.

Write own soft or use Keel ([Dear Guest/Member you can't see link before click here to register]) or other data mining software, to find rules - for example
IF (MACD > ...) AND (price grows) AND (5 EMA > 34 EMA) AND (...)
THEN in 75% of cases the price will move up (down)

Checking all combinations of 5-8 conditions out of 20 can be done in a reasonable time. Otherwise we can try genetic algorithms, distributed computing or so.
We will be rather interested in cases that occur often in the history.
Of course we can add time, day of week and news filter (indicator).

What do experienced traders think about this idea?