Ensemble and optimization algorithm in support vector machines for classification of wheat genotypes

Area of Science:

• Agricultural Science
• Computational Biology
• Machine Learning

Background:

• Accurate classification of wheat genotypes is crucial for crop improvement and breeding programs.
• Traditional methods may not fully leverage complex genotypic and phenotypic data.
• Machine learning offers advanced tools for analyzing large agricultural datasets.

Purpose of the Study:

• To classify 302 wheat genotypes using Support Vector Machines (SVMs).
• To enhance SVM classification accuracy through ensemble algorithms and optimization techniques.
• To evaluate the effectiveness of different SVM kernels and optimization methods for wheat genotype identification.

Main Methods:

• Utilized a dataset of 302 wheat genotypes and 14 morphological attributes.
• Evaluated six Support Vector Machine (SVM) kernels: linear, radial basis function (RBF), sigmoid, and polynomial (degrees 1-3).
• Applied optimization techniques including grid search, random search, genetic algorithms, differential evolution, and particle swarm optimization (PSO).
• Employed weighted accuracy ensemble methods to further improve classification performance.

Main Results:

• The Radial Basis Function (RBF) kernel achieved the highest initial accuracy of 93.2%.
• Ensemble methods, specifically weighted accuracy ensemble, improved performance to 94.9%.
• Optimization-based SVM classification, particularly with Particle Swarm Optimization (PSO), yielded a significant 1.7% accuracy gain on the test set, reaching 94.9%.

Conclusions:

• Support Vector Machines (SVMs), especially with RBF kernels and optimization techniques like PSO, are highly effective for wheat genotype classification.
• These computational approaches significantly enhance accuracy in agricultural research.
• The findings demonstrate the potential of advanced machine learning for accelerating crop improvement and breeding efforts.