Abstract:
Support vector machines (SVMs) have been applied to many real world datasets with great success. They are widely considered state of the art methods to solve many problems. However, the leap is when they have been adopted by deep neural networks. SVMs perform very well, and can outperform the end to end neural networks when used in combination with deep neural network extracted features. This is the line we intend to investigate in this research. Different techniques for SVMs are important in order to achieve high classification accuracy, particularly with the high generalisation attributed to SVMs. We propose a novel ensemble technique for classifying multiclass hyperspectral honey dataset. This method uses an ensemble of tree-based multiclass SVMs to try reducing the effects of error propagation down the trees. The results show that this approach is comparable to the commonly used parallel method for multiclass SVM.