Least squares support vector machine ensemble based on sampling for classification of quality local cattle

Abstract

Selection of superior quality local cattle with quality meat with low water and fat content that is very suitable for food processing and supports local wisdom culture. The main problem in selecting superior quality cattle is to choose the right candidate for the parent for breeding with characteristics almost the same as non-local cattle entering Madura. Classification is done to find the best model for selecting superior seeds with unbalanced classes. Using cattle data, this study will apply the LS-SVM ensemble method with combined SMOTE for multi-class imbalanced classification. To overcome high dimensions with unbalanced classes, the gradient Boosting method and sampling technique with SMOTE are applied to balance the number of majority classes into minority classes. The evaluation criteria for classification performance use accuracy values, such as G-means and running time. The experiment used k-fold cross-validation with k=5, with ensemble gradient boosting optimization showing success in improving classification performance. While using kernels, linear kernels produce higher performance and shorter computing time, with the addition of the gradient boosting technique and the best parameters of a σ value of 10 and C value of 50, and the SMOTE sampling technique produces the highest accuracy of 100%. The addition of gradient boosting has reduced iterations to make faster time on the LS-SVM method, and the correct parameters have produced a Grid Search performance.