A recommendation system for first-order nearly orthogonal-and-balanced (NOAB) designs

The construction of nearly orthogonal-and-balanced (NOAB) designs is examined for full first-order models in the framework of an algorithm selection problem, allowing for the examination of experimental design performance measures for various design sizes and maximum allowed imbalance settings. Based on a randomly-generated set of large design spaces, performances measures of D-criterion for good parameter estimation as well as estimated maximum unscaled prediction variance (UPV) are largely driven by choice of design size, with specific design space features found to impact the measures. In this multi-objective setting, prediction of design performance within the framework consistently results in designs that perform well over an entire sampled weight space for the multiple performance measures as well as for specific weights.