Rice performance prediction to deficit irrigation using microsatellite alleles and artificial intelligence

Abstract

Rice germplasm investigated as completely randomized design under flooding and deficit irrigation conditions. The results of the association analysis indicated that RM29, RM63, and RM53 could be used for rice breeding programs to improve yields under deficit irrigation. The highest accuracy of rice performance prediction was 98.36 for the RFA (RFA) for panicle length, flag leaf length, and width, and the number of primary branches, after that, the MLP algorithm had better prediction power than other algorithms. When a genotypes code was considered as a criterion to classify the genotypes under the drought stress at the reproductive stage, the random forest algorithm (RFA) was the best algorithm based on the predictive accuracy (67.93), kappa value (0.514) and root mean square error (0.293). Based on the artificial intelligence methods, the RFA presented the best results to predict the response of genotypes to deficit irrigation using the microsatellite molecular data.