摘要
为解决不平衡大数据分类问题,提出一种基于随机森林(RF)优化的深度置信网络(DBN)模型。研究思路是先对原始数据进行预处理,再使用RF算法优化DBN超参数构建三层隐藏层的分类模型。在训练阶段,先在无监督预训练情况下用自动学习数据特征表示,然后在有监督情况下微调提高分类性能,最后在公开数据集上进行验证。结果显示,RF-DBN模型在精确率、召回率等指标上均有提升,尤其在处理高度不平衡和高维异常检测任务时表现优异。通过混淆矩阵和性能对比进一步验证了模型的有效性。未来将探索更深层的DBN变体网络,优化数据采样策略以提高模型的分类性能和鲁棒性。
This paper proposes a Random Forest(RF)optimized Deep Belief Network(DBN)model for addressing imbalanced large data classification problems.The research approach involves pre-processing the raw data,followed by using the RF algorithm to optimize the hyperparameters of the DBN and constructing a three-layer hidden classification model.During the training phase,unsupervised pre-training is first used to automatically learn data feature representations,which is then followed by supervised fine-tuning to improve classification performance.The model is validated on publicly available datasets.The results show that the RF-DBN model achieves improvements in metrics such as precision and recall,and performs particularly well in handling highly imbalanced and high-dimensional anomaly detection tasks.The effectiveness of the model is further verified through confusion matrix analysis and performance comparisons.It provides insight for exploring deeper DBN variant networks and optimized data sampling strategies to further enhance classification performance and model robustness.
作者
胡晶
HU Jing(Fujian Chuanzheng Communications College,Fuzhou 350007,China)
出处
《广东水利电力职业技术学院学报》
2025年第3期61-65,共5页
Journal of Guangdong Polytechnic of Water Resources and Electric Engineering
基金
2023年新一代信息技术创新项目(2023IT013)
科教发展基金项目(Z202311033)。
作者简介
胡晶,女,副教授,研究方向为数据挖掘分析、大数据技术。
