you can define binary coefficients for your input vectors (features). The artificial bee colony algorithm defines which coefficient are ones and which are zeros using your cost function which is usually Root mean square error (RSME) . In case you use Matlab, Faradars tutorials explain it using Matlab codes.
In my opinion, the problem of using intelligent methods like ABC, GA, Ant colony is the high time complexity of the fitness function. In many literature, the dependency are employed as the fitness function, which will be O(|U|^2).