A New Fast Perceptron Algorithm

Classification is a popular research area of machine learning. This paper focuses on classification and devotes itself to accelerate the perceptron algorithm, which is a famous classification algorithm. As the high time cost is one main drawback of the perceptron algorithm, the proposed algorithm adopts a hyperplane fitting strategy to find a middle hyperplane between the positive and negative sample hyperplanes. Then, it takes the normal vector and constant of the middle hyperplane as the initial parameters for the perceptron. This strategy can fast accelerate the convergence of the perceptron. For evaluating the performance of the proposed method, experiments on real datasets Iris and I-880 are performed. The perceptron algorithm is also adopted to perform experiments for comparison. Experimental results show that the classification performance of the proposed algorithm is as good as the perceptron algorithm. More importantly, the proposed algorithm has much smaller iterations than the perceptron algorithm.