Preliminary Evaluation of Search Space Characterization and Effects of Machine Learning

Abstract

The bulk of published literature in the field of Machine Learning (ML) documents variations on standard ML algorithms, or describes the application of a ML algorithm to a given problem. The current growing fad of applying ML to every problem is resulting in poor solutions and wasted resources. The optimization of the solution algorithms largely focuses upon industrial usages of ML, but rarely are effects of the underlying problem upon the ML algorithm discussed. That is, there is typically no investigation or definition on the solution space in which the ML algorithm is operating in. As there is no widespread discussion of the different forms solution spaces may take, there exists no standard for knowing when to apply ML algorithms to a given problem. This study is designed to research current terminology and characterization of search spaces and their effects upon ML algorithms by rigorous evaluation of ML performance. The result of this study shall generate an understanding on the value and feasibility of search space characteristics. A guide for search space identification, solution algorithm seeding, and expected results for baseline and seeded implementations of ML is predicted to materialize.