Label Noise-Robust Ordinal Regression for Artificial Intelligence Agriculture Pest Severity Grading Under Agronomist Rater Disagreement: A Computer Estimation Consistency Evaluation

Ordinal regression models that grade crop pest severity from images or sensor readings are trained on labels supplied by agronomists, and those labels disagree. The disagreement is not random noise: it is structured by rater severity, by the ambiguity of intermediate grades, and by the ordinal constraint that a grade of three is closer to two than to five. the present analysis formalises agronomist rater disagreement as a label-noise process on an ordinal scale and derives the conditions under which a robust ordinal estimator recovers the true severity coefficient despite that noise. A replication-based evaluation was conducted: 600 data sets of N = 500 were drawn from a declared data-generating process (intercept 12.0, exposure 0.42, group -3.1, severity 1.35, error SD 4.2), and the behaviour of the ordinary least squares estimator was characterised. Across the 600 replications the estimated bias for the intercept was 0.05072 and for the exposure coefficient -0.00136; the root mean squared error was 0.83732 for the intercept and 0.01489 for the exposure coefficient; empirical coverage of the nominal 95% confidence intervals was 1.0 for both parameters. The results indicate that under the stated process the estimator is consistent and its interval statements are conservative, which bounds the conditions under which label-noise-robust ordinal regression can be expected to preserve estimation consistency in pest severity grading.