Design of big data classifier optimization method based on improved particle swarm optimization algorithm

Abstract

Data present complex properties such as high dimension, non-linear, and large-scale. How to perform feature selection more efficiently has become the researching focus in data mining, machine learning, and other fields. Various optimization methods have been proposed, and the current big data classifier optimization has achieved certain results. However, the classification accuracy and stability of classifiers in current research have not reached the ideal state. To address the above issues, this research proposed an optimization method of big data classifier based on improved particle swarm optimization algorithm. The particle swarm optimization algorithm was first improved by dynamically adjusting inertia weights and using multiple swarm optimization methods. Then the support vector machine big data classifier was optimized by improving the particle swarm optimization algorithm. In the experimental results, the average classification accuracy of the optimized classifier was 82.23%, while the average accuracy of the traditional support vector machine and the traditional particle swarm optimization-support vector machine was 69.45% and 72.34%, respectively. In addition, although the running time of the optimized classifier was slightly higher than traditional support vector machines, the increase was not significant. When the features number was 100, the running time increased by 0.87 seconds. The results verified that the optimized classifier performed better and did not cause significant computational load while improving accuracy. The research aims to provide a reference scheme for the optimization of big data classifiers, so as to provide guidance for more efficient big data processing.