Ensemble learning is a class of machine learning methods that make decisions by constructing and combining the outputs of multiple base classifiers.Among these, the representative Bagging ensemble method is frequently utilized for classification, regression, modeling in noisy environments, and learning from imbalanced data.Addressing the issues of unstable classification performance and insufficient focus on hard-to-classify samples inherent in traditional Bagging methods when applied to noisy datasets, a new ensemble learning method named eNewBagging is proposed, which integrates Instance Hardness (IH) with an enhanced bootstrapping mechanism.Based on the eBagging framework, an IH-based weighted sampling strategy is introduced.By allocating adaptive weights to samples, this method suppresses the influence of noisy samples and intensifies the learning of hard-to-classify samples, thereby improving overall accuracy and robustness while maintaining classifier diversity.Experiments are conducted on seven public datasets from UCI and KEEL.Evaluations on the original data are performed using two base classifiers: k-Nearest Neighbors (kNN) and Decision Trees (DT).Furthermore, performance comparisons are conducted exclusively using the kNN base classifier under varying noise level ratios ranging from 2% to 30%.The results indicate that eNewBagging can achieve optimal or sub-optimal performance across the ACC, AUC, and F1 metrics.In comparison with BaggingIH and GrpMixBag, eNewBagging exhibits a smaller margin of decline on highly noisy datasets, demonstrating stronger noise robustness.While preserving the bias-variance balance, this method achieves enhanced learning of borderline samples, significantly elevating the model's generalization performance in complex data environments.It is expected that the eNewBagging method can provide an efficient framework to overcome the performance degradation of ensemble learning caused by noise interference and sample complexity.
在分类任务中,最直观的性能度量指标为准确率(Accuracy, ACC)[21],即正确预测样本占总样本的比例。然而,准确率在处理类别不平衡问题时存在一定局限[22]。因此,本研究同时采用以下3个性能指标进行综合评估:ACC用于衡量整体分类正确率,适用于平衡数据集;AUC(Area Under the ROC Curve)[23]通过计算ROC曲线下的面积评估模型区分正负类的能力,适用于不平衡数据集;F1值(F1-score)[24]结合查准率(Precision)与查全率(Recall),作为综合性指标能更全面地反映模型的分类性能。eNewBagging在算法层面支持多分类任务,其性能在多分类场景下的验证将作为未来工作的重要研究方向。
WolpertD H, MacreadyW G.No free lunch theorems for optimization [J].IEEE Trans Evol Comput, 1997, 1(1): 67-82.
[2]
KilimciZ, OmurcaS.Enhancement of the heuristic optimization based extended space forests with classifier ensembles [J].Int Arab J Inf Technol, 2020, 17(2): 188-195.
[3]
MohamedH, NegmA, ZahranM, et al.Assessment of ensemble classifiers using the bagging technique for improved land cover classification of multispectral satellite images [J].Int Arab J Inf Technol, 2018, 15(2): 270-277.
[4]
QuinlanJ R.Bagging, boosting, and C4.5 [C]// Proceedings of the National Conference on Artificial Intelligence.Menlo Park: AAAI Press, 1996: 725-730.
KohaviR, WolpertD H.Bias plus variance decomposition for zero-one loss functions [C]// International Conference on Machine Learning.Bari: IMLS, 1996: 275-283.
[7]
LiuY, LiuJ, TanS, et al.A bagging-based surrogate-assisted evolutionary algorithm for expensive multi-objective optimization [J].Neural Comput Appl, 2022, 34(14): 12097-12118.
[8]
BauerE, KohaviR.An empirical comparison of voting classification algorithms: Bagging, boosting, and variants [J].Mach Learn, 1999, 36(1): 105-139.
[9]
KimJ, ChoiS.On uncertainty estimation by tree-based surrogate models in sequential model-based optimization [C]// International Conference on Artificial Intelligence and Statistics.Brookline: PMLR, 2022: 4359-4375.
[10]
WangL, DengR, ZhangL, et al.A novel surrogate-assisted multi-objective well control parameter optimization method based on selective ensembles [J].Processes, 2024, 12(10): 2140.
[11]
WalmsleyF N, CavalcantiG D C, OliveiraD V R, et al.An ensemble generation method based on instance hardness [C]// International Joint Conference on Neural Networks.Piscataway: IEEE, 2018: 1-8.
[12]
KabirA, RuizC, AlvarezS A.Mixed bagging: A novel ensemble learning framework for supervised classification based on instance hardness [C]// International Conference on Data Mining.Piscataway: IEEE, 2018: 1073-1078.
[13]
TüysüzoğluG, BirantD。Enhanced bagging (eBagging): A novel approach for ensemble learning [J].Int Arab J Inf Technol, 2020, 17(4): 515-528.
[14]
PaivaP Y A, MorenoC C, Smith-MilesK, et al.Relating instance hardness to classification performance in a dataset: A visual approach [J].Mach Learn, 2022, 111(8): 3085-3123.
[15]
SmithM R, MartinezT, Giraud-CarrierC.An instance level analysis of data complexity [J].Mach Learn, 2014, 95(2): 225-256.
[16]
DietterichT G.Ensemble methods in machine learning [C]//International workshop on multiple classifier systems.Berlin/Heidelberg: Springer, 2000: 1-15.
[17]
DietterichT G.An experimental comparison of three methods for constructing ensembles of decision trees: Bagging, boosting, and randomization [J].Mach Learn, 2000, 40(2): 139-157.
[18]
KrawczykB, GalarM, JeleńŁ, et al.Evolutionary undersampling boosting for imbalanced classification of breast cancer malignancy [J].Appl Soft Comput, 2016, 38: 714-726.
FatourechiM, WardR K, MasonS G, et al.Comparison of evaluation metrics in classification applications with imbalanced datasets [C]// International Conference on Machine Learning and Applications.Piscataway: IEEE, 2008: 777-782.
[23]
PowersD M W.Evaluation: From precision, recall and F-measure to ROC, informedness, markedness and correlation [J].J Mach Learn Tech, 2011, 2(1): 37-63.
[24]
BekkarM, DjemaaH K, AlitoucheT A.Evaluation measures for models assessment over imbalanced data sets [J].J Inf Eng Appl, 2013, 3(10): 27-38.
[25]
KohaviR.A study of cross-validation and bootstrap for accuracy estimation and model selection [C]// International Joint Conference on Artificial Intelligence.Montreal: Morgan Kaufmann, 1995: 1137-1145.