Wheat Seed Classification Using Gurobi Optimized Piecewise Linear Approximation-Based SVM

Due to the variety of grains and labor-intensive manual sorting processes, the demand for automated grain classification is rising. Accurate grain classification is crucial for maintaining grain quality in the supply chain as distinguishing grain varieties is becoming essential. This research presents an innovative approach for efficient wheat seed classification. The proposed method works on primal optimization problem using piecewise linear approximation techniques. For implementation, the paper utilizes the wheat seed data from the UCI repository. Exploratory data analysis and feature engineering are performed on the datasets to improve the accuracy and efficiency of wheat variety categorization. Implementation is carried out using MATLAB 13 and the Gurobi optimizer, followed by validation through tenfold cross-validation on the wheat seed dataset. Comparative analysis with various machine learning classifiers, including KNN, CART, Random Forest, and SVM, demonstrates that PLASVM outperforms them all, attaining 98.3% accuracy in just 2.672 s, while SVM achieves 97.4% accuracy in 7.34 s.