Method for Improving the Reliability of Learning Machine

Optimization of learning machine prediction has been attempted by the addition of a “width” parameter proposed by Wangen et al. Predictions were made for the production of phenazine compounds, the utilizations of d-sorbitol, of n-hexadecane, and of histamine by pseudomonads. In all but one cases examined, the “stability” of prediction was much improved when a high positive value was assigned to the parameter. A slight increase was often found in the hit-to-miss ratio, but the χ² test revealed that it was statistically insignificant.