Oil yield prediction for sunflower hybrid selection using different machine learning algorithms.

Area of Science:

• Agricultural Science
• Plant Breeding
• Computational Biology

Background:

• Increasing global demand for sunflower oil necessitates development of high-yielding hybrids.
• Sunflower Oil Yield Prediction (SOYP) aids breeders in identifying superior hybrids using advanced techniques.
• Machine learning (ML) offers a promising approach for accurate SOYP.

Purpose of the Study:

• To develop and compare ML models for predicting sunflower oil yield.
• To identify the most relevant features for accurate SOYP.
• To evaluate the potential of ML in sunflower breeding programs.

Main Methods:

• Developed and compared four ML algorithms: Artificial Neural Network (ANN), Support Vector Regression, K-Nearest Neighbour, and Random Forest Regressor (RFR).
• Utilized a dataset of 1250 sunflower hybrids, with 70% for training and 30% for testing.
• Evaluated model performance using Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and R-squared (R2) metrics.

Main Results:

• The Random Forest Regressor (RFR) consistently outperformed other models, achieving an R2 of 0.92 in 2019.
• Artificial Neural Network (ANN) recorded the lowest MAE (65) in 2018.
• Key predictors for SOYP included seed yield, resistance to broomrape and downy mildew, maturity, and locality (influenced by weather).

Conclusions:

• ML, particularly RFR, demonstrates significant potential for sunflower oil yield prediction and genotypic selection.
• Incorporating traits like disease resistance and maturity enhances prediction accuracy.
• Locality is a valuable feature for SOYP, though its effectiveness is weather-dependent.