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

Saptak Das; Arijit Sarkar; Soumik Neogi; Suborno Naskar

cs.LGOct 27, 2026

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

Authors: Saptak Das, Arijit Sarkar, Soumik Neogi, Suborno Naskar

Organizations: Indian Institute of Science Education and Research Thiruvananthapuram, Kerala, India · Department of Information Technology, MCKV Institute of Engineering, Howrah, West Bengal, India

Abstract

Reliable crop-yield prediction can support agricultural planning and resource allocation. This study evaluates five regression models for predicting crop yield from environmental, crop-related, and agricultural management variables. The dataset initially contained 1,000,000 observations and ten variables, with 231 observations having negative target values removed during preprocessing. Categorical variables were represented using one-hot encoding, and Boolean variables were converted into binary values. The models considered were multiple linear regression (LR), random-forest regression (RF), decision-tree regression (DT), ridge regression, and lasso regression. Their performance was evaluated using 80:20 and 70:30 training-testing splits based on the coefficient of determination (R-squared), mean squared error (MSE), and root mean squared error (RMSE). Applying an 80:20 split, LR achieved a test R-squared of approximately 0.9130 and a test RMSE of 0.5008. Applying a 70:30 split, lasso regression yielded the highest test R-squared (approximately 0.9131) and the lowest test RMSE (approximately 0.5008), with performance almost identical to LR and ridge regression. DT also showed improved test performance under the 70:30 split, whereas RF exhibited a substantial gap between training and testing performance. These results indicate that the evaluated linear regression model provides competitive baselines for this dataset. Further experimentation, including validation testing, will help determine the models' performance and predictability.

Figures & tables

Variable	Role	Description
Region	Predictor	Geographic region (East, North, South, West)
Soil_Type	Predictor	Soil category (Chalky, Clay, Loam, Peaty, Sandy, Silt)
Crop	Predictor	Crop category (Barley, Cotton, Maize, Rice, Soybean, Wheat)
Rainfall_mm	Predictor	Rainfall in millimetres
Temperature_Celsius	Predictor	Temperature in degrees Celsius
Fertilizer_Used	Predictor	Boolean indicator of fertilizer use

Table 1: Variables in the supplied raw dataset.

Model	80:20 split	70:30 split
LR	20 predictors; untuned	24 predictors; untuned
RF	24 predictors; untuned	24 predictors; untuned
DT	24 predictors; random search	24 predictors; grid search
Ridge	24 predictors; untuned	24 predictors; untuned
Lasso	24 predictors; grid search	24 predictors; grid search

Table 2: Experimental configurations for the two train–test splits.

Model	Tr. $R^{2}$	Tr. MSE	Tr. RMSE	Te. $R^{2}$	Te. MSE	Te. RMSE
LR	0.9129706418	0.2504014163	0.5004012553	0.9130137709	0.2507766686	0.5007760663
RF	0.9870627444	0.03722315308	0.1929330275	0.9077351184	0.2659947428	0.5157467817
DT	0.9161022839	0.2413910359	0.4913156174	0.909994562	0.2594808871	0.5093926649
Ridge Regression	0.9129706417	0.2504014163	0.5004012553	0.9130137621	0.250776694	0.5007760917
Lasso Regression	0.9129703752	0.2504021831	0.5004020215	0.9130138109	0.2507765533	0.5007759512

Table 3: Regression performance for the 80:20 train–test split.

Model	Methodology (80:20)	Hyperparameter (If any)
LR	20 features; without hyperparameter tuning	—
RF	24 features; without hyperparameter tuning	—
DT	24 features; with hyperparameter tuning; RandomizedSearchCV	max_depth = 52 min_samples_split = 16 min_samples_leaf = 93
Ridge Regression	24 features; without hyperparameter tuning	—
Lasso Regression	24 features; with hyperparameter tuning; GridSearchCV	alpha = 0.0001

Table 4: Model configurations and hyperparameters for the 80:20 train–test split

Model	Tr. $R^{2}$	Tr. MSE	Tr. RMSE	Te. $R^{2}$	Te. MSE	Te. RMSE
LR	0.9129304560	0.2503510489	0.5003509258	0.9130916682	0.2507726187	0.5007720227
RF	0.9870488887	0.0372383284	0.1929723516	0.9077328853	0.2662353023	0.5159799437
DT	0.9134832049	0.2487617300	0.4987601929	0.9119788539	0.2539836270	0.5039678829
Ridge Regression	0.912930456	0.25035104897	0.5003509258	0.913091663	0.25077263296	0.500772036
Lasso Regression	0.9129301770	0.2503518512	0.500351727	0.9130925912	0.2507699556	0.500769363

Table 5: Regression performance for the 70:30 train–test split.

Model	Methodology (70:30)	Hyperparameter (If any)
LR	24 features; without hyperparameter tuning	—
RF	24 features; without hyperparameter tuning	—
DT	24 features; with hyperparameter tuning; GridSearchCV	max_depth = 10 min_samples_split = 2 min_samples_leaf = 4
Ridge Regression	24 features; without hyperparameter tuning	—
Lasso Regression	24 features; with hyperparameter tuning; GridSearchCV	alpha = 0.0001

Table 6: Model configurations and hyperparameters for the 70:30 train–test split

Model	Tr. $R^{2}$	Tr. MSE	Tr. RMSE	Te. $R^{2}$	Te. MSE	Te. RMSE
LR	0.9129706418	0.2504014163	0.5004012553	0.9130137709	0.2507766686	0.5007760663
RF	0.987018294	0.03735104615	0.1932641874	0.907479335	0.2667321528	0.5164611823
DT	1	0	0	0.8156732457	0.5314042218	0.7289747745
Ridge Regression	0.9129706417	0.2504014163	0.5004012553	0.9130137621	0.250776694	0.5007760917
Lasso Regression	0.8847181529	0.3304832303	0.5748767088	0.885137457	0.3323514293	0.5764992882

Table 7: Regression performance for the 80:20 train–test split without hyperparameter tuning using 20 features.

Model	Tr. $R^{2}$	Tr. MSE	Tr. RMSE	Te. $R^{2}$	Te. MSE	Te. RMSE
LR	0.9129705048	0.2504018102	0.5004016488	0.9130139399	0.2507761815	0.50077558
RF	0.9441125995	0.1607995798	0.4009982291	0.9092718421	0.2615644502	0.5114337202
DT	0.9160686443	0.2414878239	0.4914141063	0.9100197862	0.2594081671	0.5093212808
Ridge Regression	0.9129706415	0.250401417	0.500401256	0.9130137194	0.2507768169	0.5007762144
Lasso Regression	0.5846677618	1.194996525	1.093158966	0.5836824697	1.200221281	1.095546111

Table 8: Regression performance for the 80:20 train–test split using RandomizedSearchCV with 20 features.

Model	Tr. $R^{2}$	Tr. MSE	Tr. RMSE	Te. $R^{2}$	Te. MSE	Te. RMSE
LR	0.9129705048	0.2504018102	0.5004016488	0.9130139399	0.2507761815	0.50077558
RF	0.9441125995	0.1607995798	0.4009982291	0.9092718421	0.2615644502	0.5114337202
DT	0.9134173038	0.2491162775	0.4991154952	0.9120229454	0.2536331658	0.503620061
Ridge Regression	0.9129706417	0.2504014163	0.5004012553	0.9130137621	0.250776694	0.5007760917
Lasso Regression	0.9129703047	0.250402386	0.5004022242	0.9130134481	0.2507775991	0.5007769954

Table 9: Regression performance for the 80:20 train–test split using GridSearchCV with 20 features.

Model	Tr. $R^{2}$	Tr. MSE	Tr. RMSE	Te. $R^{2}$	Te. MSE	Te. RMSE
LR	0.9129706418	0.2504014163	0.5004012553	0.9130137709	0.2507766686	0.5007760663
RF	0.9870627444	0.03722315308	0.1929330275	0.9077351184	0.2659947428	0.5157467817
DT	0.583743874773	1.504864234	1.226729487	0.8151677147	0.5328616407	0.7299737261
Ridge Regression	0.9129706417	0.2504014163	0.5004012553	0.9130137621	0.250776694	0.5007760917
Lasso Regression	0.8847181529	0.3304832303	0.5748767088	0.885137457	0.3323514293	0.5764992882

Table 10: Regression performance for the 80:20 train–test split without hyperparameter tuning using 24 features.

Model	Tr. $R^{2}$	Tr. MSE	Tr. RMSE	Te. $R^{2}$	Te. MSE	Te. RMSE
LR	0.9129705515	0.2504016758	0.5004015146	0.9130136856	0.2507769147	0.500776312
RF	0.941970508	0.1669628189	0.4086108404	0.9064520849	0.26969366	0.5193203828
DT	0.9161022839	0.2413910359	0.4913156174	0.909994562	0.2594808871	0.5093926649
Ridge Regression	0.9129706415	0.250401417	0.500401256	0.9130137195	0.2507768168	0.5007762143
Lasso Regression	0.5846677618	1.194996525	1.093158966	0.5836824697	1.200221281	1.095546111

Table 11: Regression performance for the 80:20 train–test split using RandomizedSearchCV with 24 features.

Model	Tr. $R^{2}$	Tr. MSE	Tr. RMSE	Te. $R^{2}$	Te. MSE	Te. RMSE
LR	0.9129705515	0.2504016758	0.5004015146	0.9130136856	0.2507769147	0.500776312
RF	0.941970508	0.1669628189	0.4086108404	0.9064520849	0.26969366	0.5193203828
DT	0.9134186857	0.2491123014	0.499111512	0.9120196948	0.2536425372	0.5036293649
Ridge Regression	0.9129706417	0.2504014163	0.5004012553	0.9130137621	0.250776694	0.5007760917
Lasso Regression	0.9129703752	0.2504021831	0.5004020215	0.9130138109	0.2507765533	0.5007759512

Table 12: Regression performance for the 80:20 train–test split using GridSearchCV with 24 features.

Model	Tr. $R^{2}$	Tr. MSE	Tr. RMSE	Te. $R^{2}$	Te. MSE	Te. RMSE
LR	0.912930456	0.2503510489	0.5003509258	0.9130916683	0.2507726187	0.5007720227
RF	0.9870013283	0.03737507915	0.193326354	0.9074236222	0.2671276761	0.5168439572
DT	1	0	0	0.8159752718	0.531000447	0.7286977748
Ridge Regression	0.912930456	0.250351049	0.5003509258	0.9130916633	0.250772633	0.500772037
Lasso Regression	0.585175823	1.192743905	1.092128154	0.5858101289	1.195138332	1.093223825

Table 13: Regression performance for the 70:30 train–test split without hyperparameter tuning using 20 features.

Model	Tr. $R^{2}$	Tr. MSE	Tr. RMSE	Te. $R^{2}$	Te. MSE	Te. RMSE
LR	0.9129301202	0.2503520145	0.5003518907	0.9130924958	0.2507702309	0.5007696385
RF	0.9440043794	0.1610042011	0.4012532879	0.9094332743	0.2613288568	0.5112033419
DT	0.9159929172	0.2415455547	0.4914728422	0.9101203013	0.2593464514	0.509260691
Ridge Regression	0.9129304557	0.25035105	0.5003509268	0.9130916393	0.2507727024	0.5007721063
Lasso Regression	0.5842710734	1.195345331	1.093318495	0.584907467	1.195345331	1.09331818164

Table 14: Regression performance for the 70:30 train–test split using RandomizedSearchCV with 20 features.

Model	Tr. $R^{2}$	Tr. MSE	Tr. RMSE	Te. $R^{2}$	Te. MSE	Te. RMSE
LR	0.9129301202	0.2503520145	0.5003518907	0.9130924958	0.2507702309	0.5007696385
RF	0.9440043794	0.1610042011	0.4012532879	0.9094332743	0.2613288568	0.5112033419
DT	0.913482506	0.2487637397	0.4987622076	0.9119783059	0.2539852084	0.5039694518
Ridge Regression	0.912930456	0.250351049	0.5003509258	0.9130916633	0.250772633	0.500772037
Lasso Regression	0.9129301542	0.2503519168	0.5003517936	0.9130926394	0.2507698165	0.5007692248

Table 15: Regression performance for the 70:30 train–test split using GridSearchCV with 20 features.

Model	Tr. $R^{2}$	Tr. MSE	Tr. RMSE	Te. $R^{2}$	Te. MSE	Te. RMSE
LR	0.912930456	0.2503510489	0.5003509258	0.9130916683	0.2507726187	0.5007720227
RF	0.9870488888	0.03723832849	0.1929723516	0.9077328853	0.2662353024	0.5159799438
DT	0.58374388383	1.504864284	1.2267282564	0.8157054124	0.5317791221	0.7292318713
Ridge Regression	0.912930456	0.250351049	0.5003509258	0.9130916634	0.250772633	0.500772036
Lasso Regression	0.585175823	0.585776991	0.7653606934	0.5858101289	1.195138332	1.093223366

Table 16: Regression performance for the 70:30 train–test split without hyperparameter tuning using 24 features.

Model	Tr. $R^{2}$	Tr. MSE	Tr. RMSE	Te. $R^{2}$	Te. MSE	Te. RMSE
LR	0.912930186	0.2503518254	0.5003517018	0.9130922801	0.2507708534	0.5007702601
RF	0.9416447868	0.1677887372	0.4096202354	0.906353173	0.2702164405	0.5198234705
DT	0.9160351483	0.2414241278	0.4913492931	0.9100678899	0.2594976836	0.5094091515
Ridge Regression	0.9129304557	0.25035105	0.5003509268	0.9130916393	0.2507727024	0.5007721063
Lasso Regression	0.5842710734	1.195345331	1.093318495	0.584907467	1.197742949	1.094414432

Table 17: Regression performance for the 70:30 train–test split using RandomizedSearchCV with 24 features.

Model	Tr. $R^{2}$	Tr. MSE	Tr. RMSE	Te. $R^{2}$	Te. MSE	Te. RMSE
LR	0.912930186	0.2503518254	0.5003517018	0.9130922801	0.2507708534	0.5007702601
RF	0.9416447868	0.1677887372	0.4096202354	0.906353173	0.2702164405	0.5198234705
DT	0.913483205	0.2487617301	0.4987601929	0.9119788539	0.253983627	0.5039678829
Ridge Regression	0.912930456	0.250351049	0.5003509258	0.9130916634	0.250772633	0.5007720369
Lasso Regression	0.912930177	0.2503518512	0.5003517275	0.9130925912	0.2507699556	0.5007693637

Table 18: Regression performance for the 70:30 train–test split using GridSearchCV with 24 features.

Crop Yield Prediction

Crop yield prediction estimates harvests from crop, weather, soil, satellite, and management data to support agricultural planning.

See all 18 papers

YieldSAT: A Multimodal Benchmark Dataset for High-Resolution Crop Yield Prediction
Apr 1, 2026Miro Miranda, Deepak Pathak, Patrick Helber +10Crop Yield Prediction Agricultural Remote Sensing
RPTU Kaiserslautern-Landau · DFKI GmbH · Vision Impulse GmbH +1
Crop Yield Prediction for Punjab, Pakistan: A Tree-Ensemble and Leaf-Health Prototype, and What Random Validation Hides
Oct 5, 2026Amina Asif, Qurat Ul ain Asif, Noor Bakhat Asif +1Crop Yield Prediction Data Leakage
Department of Artificial Intelligence, The Islamia University of Bahawalpur, Pakistan · Key Laboratory of Xinjiang Endemic Phytomedicine Resources, Ministry of Education; School of Pharmacy, Shihezi University, Shihezi 832002, China · Department of Physics, Government College of Women University of Faisalabad, Pakistan
Retrieval-Augmented Multi-scale Framework for County-Level Crop Yield Prediction Across Large Regions
Mar 7, 2026Yiming Sun, Qi Cheng, Licheng Liu +3Crop Yield Prediction Spatiotemporal Forecasting
University of Pittsburgh, Pittsburgh, PA, USA · University of Wisconsin-Madison, Madison, WI, USA · University of Alabama, Tuscaloosa, AL, USA +1

Apr 1, 2026J/K move · Enter open · S save

YieldSAT: A Multimodal Benchmark Dataset for High-Resolution Crop Yield Prediction

Miro Miranda, Deepak Pathak, Patrick Helber, Benjamin Bischke+9

RPTU Kaiserslautern-Landau · DFKI GmbH · Vision Impulse GmbH+1

Crop Yield Prediction Agricultural Remote Sensing Remote Sensing Benchmarks Multimodal Fusion

Open paper AskPDF arXiv

Crop yield prediction requires substantial data to train scalable models. However, creating yield prediction datasets is constrained by high acquisition costs, heterogeneous data quality, and data privacy regulations. Consequently, existing datasets are scarce, low in quality, or limited to regional levels or single crop types, hindering the development of scalable data-driven solutions. In this work, we release YieldSAT, a large, high-quality, and multimodal dataset for high-resolution crop yield prediction. YieldSAT spans various climate zones across multiple countries, including Argentina, Brazil, Uruguay, and Germany, and includes major crop types, including corn, rapeseed, soybeans, and wheat, across 2,173 expert-curated fields. In total, over 12.2 million yield samples are available, each with a spatial resolution of 10 m. Each field is paired with multispectral satellite imagery, resulting in 113,555 labeled satellite images, complemented by auxiliary environmental data. We demonstrate the potential of large-scale and high-resolution crop yield prediction as a pixel regression task by comparing various deep learning models and data fusion architectures. Furthermore, we highlight open challenges arising from severe distribution shifts in the ground truth data under real-world conditions. To mitigate this, we explore a domain-informed Deep Ensemble approach that exhibits significant performance gains. The dataset is available at https://yieldsat.github.io/.

Abstract

Crop Yield Prediction

Explore similar work

YieldSAT: A Multimodal Benchmark Dataset for High-Resolution Crop Yield Prediction

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

Retrieval-Augmented Multi-scale Framework for County-Level Crop Yield Prediction Across Large Regions