IEEE Access (Jan 2024)

Optimization Strategy of Classification Model Based on Weighted Implicit Optimal Extreme Learning Machine

  • Na Zhang,
  • Xiaofeng Wang

DOI
https://doi.org/10.1109/ACCESS.2024.3404606
Journal volume & issue
Vol. 12
pp. 74169 – 74184

Abstract

Read online

Classification algorithms are one of the important research topics in the artificial intelligence, widely applied in various scientific and engineering fields. Extreme learning machine is a single hidden layer feed-forward neural network algorithm. Compared with traditional neural network models, the training speed and the generalization ability are also better. In terms of methodology, this study first innovatively improves the traditional Grey Wolf Optimization (GWO) algorithm to enhance its convergence and search ability. Specific improvement measures include implementing the reverse learning strategy to reduce the initial dependence of the algorithm on population distribution, and adding exploration perception strategy to enhance its global search ability by calculating heuristic factors, so as to identify the global optimal solution more effectively. The results showed that the improved W-DH-ELM model had excellent performance on multiple standard data sets. In particular, the average accuracy was more than 90%, which was significantly higher than other benchmark classification models. In terms of operation efficiency, the running time of the new model on different data sets was significantly reduced, accounting for less than 25%, and the lowest running time was only 4.89%. These experimental results verify the effectiveness of the introduced intelligent optimization algorithm in improving the performance of traditional ELM model without changing the original model structure. The improved W-DH-ELM model not only maintains the fast training performance of ELM, but also has higher accuracy and stability, which shows its superiority in dealing with complex classification tasks. In summary, the weighted two-hidden layer extreme learning machine optimized by the improved GWO proposed in this study has significant advantages in classification problems, providing a new perspective for future machine learning applications and research.

Keywords