During the study of mevinolin biosynthesis by Aspergillus terreus ATCC
20542, 10 different medium components were selected for medium optimi
zation. A new optimization method based on genetic algorithms and indu
ctive learning was used for experimental design. For better efficiency
the method was supported by a model, constructed with machine learnin
g method, to predict the productivity. In four generations of fermenta
tion experiments the productivity increased nearly three times.