Abstract: Rice ragged stunt virus (RRSV) significantly impacts rice production across Asia and other major rice-growing regions. The virus infects rice during the vulnerable seedling stage, resulting in severe yield losses. Developing reliable assessment models for RRSV infection is therefore essential for early management and offers a cost-effective strategy for disease control. In this study, we constructed a machine learning model using linear regression, decision tree, and random forest algorithms based on reverse transcription quantitative polymerase chain reaction (RT-qPCR) data from pooled seedling samples of 10, 20, 30, and 50 individuals, respectively. Model performance was comprehensively evaluated based on coefficient of determination (R^?), mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), symmetric mean absolute percentage error (SMAPE), and mean absolute percentage error (MAPE), alongside comparisons of predicted and actual infection rates determined by PCR and RT-qPCR, as well as disease incidence observed in field validation assays. The model derived from 30-seedling bulks achieved the highest predictive accuracy, with estimates closely matching actual infection rates and observed field outcomes. A single RT-qPCR assay of 30 pooled seedlings thus provides a reliable and efficient means for RRSV infection assessment and disease incidence prediction under field conditions. This approach supports decision-making for seedling transplantation and enables monitoring of disease dynamics across developmental stages, ultimately facilitating timely interventions to mitigate yield losses and economic burdens on farmers.

Similar articles