基于蚁群优化的极限学习机选择性集成学习算法

Selective Ensemble Learning Algorithm of Extreme Learning Machine Based on Ant Colony Optimization

针对现有极限学习机集成学习算法分类精度低、泛化能力差等缺点,提出了一种基于蚁群优化思想的极限学习机选择性集成学习算法。该算法首先通过随机分配隐层输入权重和偏置的方法生成大量差异的极限学习机分类器,然后利用一个二叉蚁群优化搜索算法迭代地搜寻最优分类器组合,最终使用该组合分类测试样本。通过12个标准数据集对该算法进行了测试,该算法在9个数据集上获得了最优结果,在另3个数据集上获得了次优结果。采用该算法可显著提高分类精度与泛化性能。

This paper proposed a novel selective ensemble learning algorithm of extreme learning machine （ELM） based on the idea of ant colony optimizatiorL The algorithm can overcome the drawbacks of the existing ensemble learning al- gorithms of ELM, such as low classification accuracy and generalization ability. Firstly, the proposed algorithm gene- rates lots of ELM classifiers by the strategy of randomly assigning input weights and biases of the hidden layer. It then uses a binary ant colony optimization algorithm to search the optimal combination of ELMs. At last, it uses the extracted combination of classifiers to classify test instances. The experimental results on 12 baseline data sets show that the pro- posed algorithm has acquired the best performance on nine data sets and the second best performance on the three re- maining data sets. Adopting the proposed algorithm can obviously help to improve the classification accuracy and gene- ralization ability.