山西农业科学2025，Vol.53Issue(5)：8-20,13.DOI:10.26942/j.cnki.issn.1002-2481.2025.05.02
基于表型性状的玉米穗质量预测模型
Phenotype-Based Prediction Model for Maize Ear Weight
摘要
Abstract
To develop an accurate prediction model for maize ear weight,and provide a certain level of decision support for high-yield maize breeding and precision field management,in this study,32 phenotypic(agronomic)traits-covering four categories(flowering time,plant morphology,ear-leaf traits,and ear characteristics)-were measured across 281 maize recombinant inbred lines.Multiple linear regression(MLR),random forest(RF),support vector machine(SVM),and artificial neural networks(ANN)were employed to construct a prediction model for maize ear weight.Models were optimized through hyper-parameter tuning under different training-testing splits and evaluated for their performance.An ensemble-mean model was subsequently generated to integrate the individual predictions,which was also applied for analyzing the key contributing factors in maize ear-weight predictions.The results showed that model performance varied after hyper-parameter tuning:RF and SVM remained robust between training and test sets,whereas ANN relied on sample size,which was prone to over-fitting at smaller training sizes(50％-70％)but showed markedly reduced errors at an 80％of training fraction.Although MLR generally achieved lower goodness-of-fit than the non-linear models,it still delivered reasonable predictive power.Across all splits,the ensemble-mean model outperformed or matched the best single model.With an 80％training fraction,the ensemble-mean model based on ear traits-where grain weight was the dominant contributing factor-achieved the best performance(R2=0.95;RMSE=7.6 g;Relative RMSE=5％).Models constructed from plant morphology traits(e.g.,plant height,tassel length,ear height)showed moderate accuracy(R2 is about 0.34),but could support early-stage prediction of ear weight.Models relying solely on flowering or ear-leaf traits had limited predictive ability(R2≤0.11).关键词
玉米/穗质量预测/农艺性状/机器学习/超参数调优/多模型集成Key words
maize/ear weight prediction/agronomic traits/machine learning/hyper-parameter tuning/model ensemble分类
农业科技引用本文复制引用
蒋艳玲,杨峰,石江,唐善勇,李申斐,胡方乾,谭先明,夏超,兰海,杨宸尧..基于表型性状的玉米穗质量预测模型[J].山西农业科学,2025,53(5):8-20,13.基金项目
国家重点研发计划(2023YFD1201100) （2023YFD1201100）
中国博士后科学基金(2024M762265) （2024M762265）
四川省科技厅国际港澳台科技创新合作项目(2025YFHZ0140) （2025YFHZ0140）
成都国家农高区智慧农业综合服务平台研究及应用项目(2023-YF08-00003-SN) （2023-YF08-00003-SN）
国家级大学生创新创业训练计划项目(202310626042) （202310626042）
