← Back to all news
arXiv Computer Vision By Amina Asif, Qurat ul ain Asif, Noor Bakhat Asif

Crop Yield Prediction for Punjab, Pakistan: A Tree-Ensemble and Leaf-Health Prototype, and What Random Validation Hides

Read the original on arXiv Computer Vision →

The paper presents a crop‑yield prediction prototype for Punjab, Pakistan that combines several machine‑learning models (Random Forest, XGBoost, SVR, Ridge‑stacked ensemble) with a MobileNetV2 leaf‑health classifier, and deploys it as a web app with SHAP explanations. An initial 80/20 split on a small merged Kaggle table yields an R² of 0.991, but an audit reveals severe data duplication, reducing performance to –0.20 when whole years are held out. Further analysis on deduplicated data and a longer FAOSTAT table shows that a simple per‑crop linear trend outperforms the complex models, and that neither pesticide use nor national temperature changes explain the residuals; the leaf‑health component, though highly accurate on image data, is not linked to yield records and adds no predictive value. whyItMatters":"The study highlights how data quality and proper validation can dramatically alter the perceived effectiveness of advanced predictive models in agricultural forecasting."

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Computer Vision.

Related stories

arXiv AI
Oct 28

Crop Yield Prediction Using Regression Models: A Comparative Evaluation of Model Performance

The study compares five regression models—multiple linear regression, random‑forest regression, decision‑tree regression, ridge regression, and lasso regression—for predicting crop yield using a dataset of 1,000,000 observations and ten variables. After preprocessing, the models were evaluated with 80:20 and 70:30 training‑testing splits, using R‑squared, MSE, and RMSE as metrics. Lasso regression and multiple linear regression achieved the highest test R‑squared (~0.913) and lowest RMSE (~0.501), with linear models providing competitive baselines and random forest showing a large training‑testing gap.

By Saptak Das, Arijit Sarkar, Soumik Neogi, Suborno Naskar
More like this →
arXiv Machine Learning
Aug 31

Accurate prediction is not profitable advice: profit-based evaluation of machine learning nitrogen recommendations in winter wheat

The study evaluates machine learning models for nitrogen recommendations in winter wheat by directly scoring profit loss on 892 yield response curves, rather than relying on prediction accuracy. Results show that none of the models recover the best rate within farm tolerance, and at typical prices all models underperform standard UK advice. A simple post‑model correction step reduces profit losses by up to 43% without retraining, while a hybrid approach further mitigates bias and large losses.

By Xulong Wang, Po Yang
benchmarkssafety
More like this →
arXiv Machine Learning
Aug 28

Profit based evaluation of machine learning for nitrogen recommendations in winter wheat

The study evaluates machine learning for nitrogen recommendations in winter wheat by directly scoring profit loss on 892 yield response curves, rather than relying on prediction accuracy. Results show that ML models alone fail to recover the best nitrogen rate within farm tolerance and underperform standard UK advice across price scenarios. However, a simple post‑model correction step significantly reduces profit losses, and a hybrid approach combining standard advice with a damped correction eliminates bias and large losses.

By Xulong Wang, Po Yang
benchmarkssafety
More like this →