Prediction method of sugarcane important phenotype data based on multi-model and multi-task.

Area of Science:

• Agricultural Science
• Biotechnology
• Data Science

Background:

• Global food security relies on accurate sugarcane yield prediction models.
• Machine learning (ML) algorithms offer superior precision compared to remote sensing for yield forecasting.
• Phenotypic traits are crucial determinants of sugarcane yield.

Purpose of the Study:

• To develop an intelligent model ensemble for predicting sugarcane stem diameter and plant height.
• To evaluate the performance of eight ML algorithms for sugarcane yield prediction.
• To enhance overall sugarcane yield through accurate phenotypic trait prediction.

Main Methods:

• Utilized six key sugarcane phenotypic traits: plant height, stem diameter, internode length, leaf length, leaf width, and field brix.
• Applied eight ML algorithms: logistic regression, linear regression, K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Backpropagation Neural Network (BPNN), Decision Tree, Random Forest, and XGBoost.
• Developed an intelligent model ensemble for predicting stem diameter and plant height.

Main Results:

• The XGBoost algorithm demonstrated superior performance in predicting sugarcane phenotypic traits compared to seven other algorithms.
• XGBoost exhibited enhanced stability in specialized data environments with self-prepared data.
• Phenotypic trait data significantly impacted the efficacy of the intelligent prediction models.

Conclusions:

• A sugarcane yield prediction model ensemble using ML algorithms can accurately forecast stem diameter and plant height.
• This approach provides a valuable reference for artificial breeding of new sugarcane varieties with improved stem diameter and plant height.
• The study bridges a research gap in indirect yield prediction using sugarcane phenotypic traits.