Answer: Yes – in a very simple form
Machine learning of the kind used in EvalC3 has these elements:
- A search function that finds a set of attributes that MAY be a good predictor of the outcome of interest
- An evaluation function, that tests the performance of this possible predictor
- A memory function, that stores the result or any previous result, whichever was the better performer.
- Reiteration of the above process (1-3)
- A stopping rule, which says when to stop the search and evaluate functions and to publish the best performing predictors identified so far.
Machine learning is all about incremental search and progressive improvement of candidate solutions (predictive models)
Some forms of machine learning are more sophisticated than others in how they do this. For example, genetic algorithms (built into the Solver add-in used in EvalC3) don’t systematically search all possible combinations (which can take a lot of time). Instead, they search for and test variations of previously tested models, using a combination of random variation and recombination.