machine learning - How to optimized input parameters? -

i looking right machine approach optimize 5 program parameters. have growing table past results.

which machine learning method serve in scenario?

edit:

the training data simple table:

time | p1 | p2 | p3 | p4 | time elapsed

there 200k rows , steady growing. optimization minimize "time elapsed" dependent on other factors parameters parameters biggest influence in every sample taken.

i looking recommendable types of neural networks (or other methods). thought propagation or maybe pso lack experience (only ever coded ga) decide start.

yes, that's big help. note stackoverflow defined place specific questions. however, one, think, straightforward enough.

i don't know how might be; i'm trying start simply.

you have large enough data base build prediction model; i'd start logistic regression , see how in predicting own training data. have ann software lets "seed" learning such model? best of both worlds.

if incoming data dominate initial data (say, growing @ least 10x), suggest neural network, if have start scratch. output depend on each input continuously? in other words, small change in input make small change in output -- or there large leaps @ data values? if changes continuous, thought of propagation should work best.

do have choice of computation methods? classic gradient descent place start. if isn't flexible enough actual data, can try moving logistic or gaussian.

have decided on error function (cost function) training? least square classic choice, appropriate function depends on desired training "shape": have fit "normal" cases great precision? need particularly smooth brane (hyperplane)? can tolerate being farther off outliers?

fortunately, have large amount of data handful of inputs; gives lot of room experiment. don't afraid add terms in experiments. instance, can load down gradient descent lot of variations each variable: x, x^2, x^3, sqrt(x), ln(x), atan(x), etc.