Objective In a big data environment, data is continuously and dynamically updated, which imposes significant limitations and challenges on traditional machine learning algorithms. Incremental learning is a learning paradigm that focuses exclusively on newly changed data based on the learning outcomes of existing models, substantially improving learning efficiency during data update processes. Incremental attribute reduction is a widely adopted method and strategy for achieving efficient attribute reduction in dynamic dataset environments. However, in ordered information systems with dynamically updated and evolving objects, existing incremental attribute reduction methods neglect the potential classification information provided by rough approximation sets, which limits their effectiveness in supporting attribute reduction. Methods In existing research results, scholars jointly proposed a relative decision entropy model by utilizing the roughness and dependency of upper and lower approximation sets to combine information entropy. This entropy model considered both the classification information of the upper and lower approximation sets of attributes and the classification information from the perspective of attribute information entropy, and it demonstrated better performance in attribute uncertainty measurement and attribute reduction. At the same time, the basic calculation unit of relative decision entropy was the upper and lower approximation sets of attributes. When the object changed, there was no need to calculate the information granules of the updated object. Therefore, this study extended relative decision entropy to mixed ordered information system environments. First, the relative decision entropy model of dominance-based neighborhoods was proposed. Then, the relative decision entropy of dominance-based neighborhoods was constructed based on the dominance-based neighborhood relation. The relative decision entropy of dominance-based neighborhoods was reconstructed in the form of a matrix, and a non-incremental attribute reduction algorithm for hybrid ordered information systems was designed. Finally, for the two scenarios of increasing and decreasing objects in a hybrid ordered information system, the matrix-form incremental update of the relative decision entropy of the dominance-based neighborhood was analyzed and studied, and incremental attribute reduction algorithms were constructed using this update mechanism. Results and Discussions In the experimental stage, 8 public datasets were selected for simulation experiments to compare the incremental algorithm with non-incremental algorithms. In the case of increasing the number of objects in the dataset, the number of reduced attributes for the non-incremental and incremental algorithms was basically similar across the 8 datasets, with averages of 14.75 and 14.62 attributes, respectively. The classification accuracy of the non-incremental algorithm and the incremental algorithm was also basically similar across the 8 datasets, with average SVM classification accuracies of 86.12% and 86.65%, and average NB classification accuracies of 87.22% and 87.24%, respectively. The processing time of the incremental algorithms was significantly shorter than that of the non-incremental algorithms. The processing time of the non-incremental algorithms on the 8 datasets was 4 215.13 seconds, whereas the processing time of the incremental algorithms on the 8 datasets was only 55.38 seconds. The performance of the incremental algorithms was significantly superior. In the case of reducing dataset objects, the number of reduced attributes for the non-incremental and incremental algorithms was basically similar across the 8 datasets, with averages of 16.00 and 15.62, respectively. The classification accuracy of the non-incremental algorithm and the incremental algorithm was basically similar across the 8 datasets, with average SVM classification accuracies of 86.25% and 86.74%, and average NB classification accuracies of 87.03% and 87.48%, respectively. The processing time of the incremental algorithms was significantly shorter than that of the non-incremental algorithms, with the non-incremental algorithms consuming 1 053.56 seconds and the incremental algorithms consuming only 36.92 seconds. The effectiveness of the incremental algorithm was verified through these experimental results. An experimental comparison was conducted between the incremental algorithm proposed in this study and four superior comparative algorithms to verify the superiority of the algorithm. In the case of increasing the number of objects in the dataset, the number of reduced attributes for the four comparative algorithms and the algorithm proposed in this study were 14.76, 16.75, 14.87, 14.62, and 14.62, respectively. The SVM classification accuracies of the four comparative algorithms and the algorithm proposed in this study were 87.53%, 83.84%, 89.20%, 86.84%, and 89.69%, respectively. The NB classification accuracies of the four comparative algorithms and the algorithm proposed in this study were 86.10%, 84.51%, 88.00%, 87.44%, and 89.81%, respectively. In terms of processing time, the proposed algorithm was 35%, 32%, 40%, and 7% faster than the four compared algorithms, respectively. In the case of reducing dataset objects, the number of reduced attributes for the four comparative algorithms and the algorithm proposed in this study were 16.00, 18.37, 15.90, 15.82, and 15.62, respectively. The SVM classification accuracies of the four comparative algorithms and the algorithm proposed in this study were 85.92%, 84.06%, 87.67%, 86.74%, and 89.50%, respectively. The NB classification accuracies of the four comparative algorithms and the algorithm proposed in this study were 85.82%, 83.85%, 88.75%, 87.43%, and 90.15%, respectively. In terms of processing time, the proposed algorithm was 55%, 46%, 63%, and 12% faster than the four compared algorithms, respectively. Comparing these results demonstrated that the incremental algorithm in this study selected fewer attributes, achieved higher classification accuracy, and significantly outperformed the comparative algorithms in terms of efficiency. Conclusions The experimental results demonstrate that the proposed incremental algorithm exhibits superior attribute reduction performance on dynamic datasets, significantly enhancing the efficiency of dynamic attribute reduction while maintaining the number of attribute selections and classification accuracy. At the same time, the proposed incremental algorithm selects fewer attributes and achieves higher classification accuracy compared to similar algorithms. Most importantly, the algorithm demonstrates higher computational performance.
ChenBaoguo, ChenLei, DengMing,et al.Incremental attribute reduction algorithm based on incomplete hybrid order information system[J].Advanced Engineering Sciences,2024,56(1):65‒81. doi:10.15961/j.jsuese.202201214
QuKanglin, XuJiucheng, HouQincheng,et al.Feature selection using Information Gain and decision information in neighborhood decision system[J].Applied Soft Computing,2023,136:110100. doi:10.1016/j.asoc.2023.110100
[4]
YangXin, YangYuxuan, LuoJunfang,et al.A unified incremental updating framework of attribute reduction for two-dimensionally time-evolving data[J].Information Sciences,2022,601:287‒305. doi:10.1016/j.ins.2022.04.026
[5]
HuangYanyong, GuoKejun, YiXiuwen,et al.Matrix representation of the conditional entropy for incremental feature selection on multi-source data[J].Information Sciences,2022,591:263‒286. doi:10.1016/j.ins.2022.01.037
SangBinbin, ChenHongmei, YangLei,et al.Incremental attribute reduction approaches for ordered data with time-evolving objects[J].Knowledge-Based Systems,2021,212:106583. doi:10.1016/j.knosys.2020.106583
[12]
SangBinbin, ChenHongmei, LiTianrui,et al.Incremental approaches for heterogeneous feature selection in dynamic ordered data[J].Information Sciences,2020,541:475‒501. doi:10.1016/j.ins.2020.06.051
[13]
SangBinbin, ChenHongmei, YangLei,et al.Incremental feature selection using a conditional entropy based on fuzzy dominance neighborhood rough sets[J].IEEE Transactions on Fuzzy Systems,2022,30(6):1683‒1697. doi:10.1109/tfuzz.2021.3064686
[14]
PanYanzhou, XuWeihua, RanQinwen.An incremental approach to feature selection using the weighted dominance-based neighborhood rough sets[J].International Journal of Machine Learning and Cybernetics,2023,14(4):1217‒1233. doi:10.1007/s13042-022-01695-4
[15]
YangLei, QinKeyun, SangBinbin,et al.A novel incremental attribute reduction by using quantitative dominance-based neighborhood self-information[J].Knowledge-Based Systems,2023,261:110200. doi:10.1016/j.knosys.2022.110200
[16]
ZhangQinli, ChenYiying, ZhangGangqiang,et al.New uncertainty measurement for categorical data based on fuzzy information structures:An application in attribute reduction[J].Information Sciences,2021,580:541‒577. doi:10.1016/j.ins.2021.08.089
XingJinming, GaoCan, ZhouJie.Weighted fuzzy rough sets-based tri-training and its application to medical diagnosis[J].Applied Soft Computing,2022,124:109025. doi:10.1016/j.asoc.2022.109025
[19]
HuangHaixin, LiZhaowen, LiuFang,et al.New uncertainty measurement for hybrid data and its application in attribute reduction[J].Information Sciences,2024,664:120334. doi:10.1016/j.ins.2024.120334
[20]
ZhangChucai, LuZhengxiang, DaiJianhua.Incremental attribute reduction for dynamic fuzzy decision information systems based on fuzzy knowledge granularity[J].Information Sciences,2025,689:121467. doi:10.1016/j.ins.2024.121467
[21]
ThuyN N, WongthanavasuS.Attribute reduction with fuzzy divergence-based weighted neighborhood rough sets[J].International Journal of Approximate Reasoning,2024,173:109256. doi:10.1016/j.ijar.2024.109256
[22]
VoB K, NguyenH S.An attribute ranking method based on rough sets and interval-valued fuzzy sets[J].International Journal of Approximate Reasoning,2024,170:109215. doi:10.1016/j.ijar.2024.109215
[23]
ZhaoJie, WuDaiyang, WuJiaxin,et al.Consistency approximation:Incremental feature selection based on fuzzy rough set theory[J].Pattern Recognition,2024,155:110652. doi:10.1016/j.patcog.2024.110652
[24]
HuChengxiang, ZhangLi, HuangXiaoling,et al.Matrix-based approaches for updating three-way regions in incomplete information systems with the variation of attributes[J].Information Sciences,2023,639:119013. doi:10.1016/j.ins.2023.119013
[25]
WenHaotong, XuYi, LiangMeishe.Parallel attribute reduction algorithm for unlabeled data based on fuzzy discernibility matrix and soft deletion behavior[J].Information Sciences,2025,689:121472. doi:10.1016/j.ins.2024.121472
ZhangXiaoyan, WangJinghong, HouJianglong.Matrix-based approximation dynamic update approach to multi-granulation neighborhood rough sets for intuitionistic fuzzy ordered datasets[J].Applied Soft Computing,2024,163:111915. doi:10.1016/j.asoc.2024.111915
[28]
XieJingjing, HuBao qing, JiangHaibo.A novel method to attribute reduction based on weighted neighborhood probabilistic rough sets[J].International Journal of Approximate Reasoning,2022,144:1‒17. doi:10.1016/j.ijar.2022.01.010
[29]
ZhangTengfei, ZhangYudi, MaFumin.Improved neighborhood space based feature selection algorithm for highdimensional mixed data[J].Control and Decision,2024,39(3):929‒938. doi:10.13195/j.kzyjc.2022.0789
ZhangDi, ZhuPing.A novel approach to discretizing information systems associated with neighborhood rough sets[J].International Journal of Approximate Reasoning,2024,164:109062. doi:10.1016/j.ijar.2023.109062