artificial intelligence - Machine learning - predicting multiple variables that add up to 100% -

i working on toy models current presidential election. there 4 candidates , each win % of vote. goal predict each candidate's %.

so far have tried building data set 1 learned variable (% of vote rec'd) , several dozen dependent variables. using weka, have experimented mlp , several other learning methods. issue, once learn model vote %, predictions each candidates share of vote never adds 100%.

clearly in case necessity total % of votes received add 100%. approaching problem wrong? can improve method?

this happens interdependence of percentages don't seem embedded in equations. if writing scratch, somewhere include equation forcing v0 + v1 + v2 + v3 = 1. these called constraints , used impose underlying physics on set of equations otherwise wouldn't respect.

i don't oppose abhiieors answer however. if have model seems logical you, scaling fine option.

v0 = v0 * 100 / ( v0 + v1 + v2 + v3 ) etc.