Ensemble Kernel-NN: Machine Learning with Hybrid Features

Classification is a fundamental task in machine learning and data analysis that has practical applications in image and speech recognition, natural language processing, medical detection, and fault detection. However, determining whether data exhibits linear or nonlinear characteristics remains challenging in practice. This difficulty stems from two factors: (1) the inherent complexity of data characteristics and their representations, and (2) the computational overhead required for preprocessing methods such as parameter adjustment, data visualization, dimension reduction, and kernel function replacement. This paper proposes an Ensemble Kernel-NN algorithm that utilizes the idea of ensemble learning. It is based on the Radial Basis Function neural network and support vector machine method for bagging and boosting. This algorithm simplifies the entire machine-learning process by eliminating the need for judgment and preprocessing of dimensions and features. It can directly provide better classification prediction results than any hypothesis in an end-to-end manner, solving the linear and nonlinear feature classification problem simultaneously. We conducted detailed comparison experiments and ablation experiments to prove the effectiveness of our model and also performed a performance analysis of our baseline model.