A Novel Feature-Scheduling Aggregation Clustering Framework Based on Convolutional Neural Networks
Abstract
1. Introduction
2. Related Works
3. Proposed Method
3.1. Convolutional Neural Networks
3.2. Feature Scheduling
3.3. Feature Aggregators
- High-frequency branch: Utilizes max pooling to capture salient features and accentuate data variations/detail patterns.
- Low-frequency branch: Applies average pooling to derive global trend representations and background characteristics.
- High-frequency signals are emphasized via max pooling;
- Low-frequency components are distilled through average pooling.
4. Experiments
4.1. Datasets and Evaluation Metrics
- Wisconsin Breast Cancer Dataset: Contains 569 instances with 30 features, classified into benign/malignant categories;
- Heart Disease Dataset: Contains 303 instances containing 13 diagnostic features;
- IRIS Dataset: Contains 150 samples with 4 morphological features across three species classes.
4.2. Experimental Parameter Setting
4.3. Comparative Experiments
- (1)
- Breast Cancer Dataset Analysis
- SSE: 1.479 × 105 (lowest among compared methods);
- NMI: 0.593 (superior feature-target correlation capture, as shown in Figure 2);
- This tri-metric superiority demonstrates robust classification capability for breast cancer subtyping, particularly excelling in cluster cohesion (SSE) and label consistency (NMI).
- (2)
- Heart Disease Dataset Analysis
- SSE: 2.13 × 105 (lowest intra-cluster variance);
- NMI: 0.185 (optimal feature-label correlation capture, as shown in Figure 3).
- (3)
- IRIS Dataset Analysis
- SSE: 0.95436 (lowest intra-cluster variance);
- NMI: 0.853 (superior cluster-label alignment, as shown in Figure 4).
4.4. Ablation Study
- CNN-only: The baseline model only has the CNN module.
- No-Feature Dispatching (No-FD): The model removes the feature scheduling module while retaining the CNN and feature aggregation module.
- No-Feature Aggregator (No-FA): The model removes the feature aggregation module while retaining the CNN and feature scheduling module.
- No-AvgPool: The model removes the low-frequency aggregation branch while retaining the CNN, feature scheduling, and high-frequency aggregation branch modules.
- In all datasets, the largest performance gap between No-FA and Our Method is observed in the SSE metric.
- Except for the heart disease dataset, removing the feature aggregation module leads to a notable reduction in the NMI value.
5. Conclusions
Author Contributions
Funding
Data Availability Statement
Conflicts of Interest
References
- Shin, Y.-M.; Kim, S.-W.; Shin, W.-Y. PAGE: Prototype-Based Model-Level Explanations for Graph Neural Networks. IEEE Trans. Pattern Anal. Mach. Intell. 2024, 46, 6559–6576. [Google Scholar] [CrossRef] [PubMed]
- Zhang, H.; Wang, D.; Zhao, W.; Lu, Z.; Jiang, X. IMCSN: An improved neighborhood aggregation interaction strategy for multi-scale contrastive Siamese networks. Pattern Recognit. 2025, 158, 111052. [Google Scholar] [CrossRef]
- Deng, W.; Zhao, L.; Liao, Q.; Guo, D.; Kuang, G.; Hu, D.; Pietikainen, M.; Liu, L. Informative Feature Disentanglement for Unsupervised Domain Adaptation. IEEE Trans. Multimed. 2022, 24, 2407–2421. [Google Scholar] [CrossRef]
- Liu, D.; Zhang, C.; Song, Y.; Huang, H.; Wang, C.; Barnett, M.; Cai, W. Decompose to Adapt: Cross-Domain Object Detection Via Feature Disentanglement. IEEE Trans. Multimed. 2023, 25, 1333–1344. [Google Scholar] [CrossRef]
- Ding, Y.; Zhang, Z.; Zhao, X.; Cai, Y.; Li, S.; Deng, B.; Cai, W. Self-Supervised Locality Preserving Low-Pass Graph Convolutional Embedding for Large-Scale Hyperspectral Image Clustering. IEEE Trans. Geosci. Remote Sens. 2022, 60, 1–16. [Google Scholar] [CrossRef]
- Xue, Y.; Jin, G.; Shen, T.; Tan, L.; Wang, N.; Gao, J.; Wang, L. SmallTrack: Wavelet Pooling and Graph Enhanced Classification for UAV Small Object Tracking. IEEE Trans. Geosci. Remote Sens. 2023, 61, 1–15. [Google Scholar] [CrossRef]
- Malhi, U.S.; Zhou, J.; Yan, C.; Rasool, A.; Siddeeq, S.; Du, M. Unsupervised Deep Embedded Clustering for High-Dimensional Visual Features of Fashion Images. Appl. Sci. 2023, 13, 2828. [Google Scholar] [CrossRef]
- Zhou, W.; Song, W.; Zhang, Z.; Zhang, F.; Teng, Z.; Tian, Z. Hierarchical feature aggregation with mixed attention mechanism for single-cell RNA-seq analysis. Expert Syst. Appl. 2025, 260, 125340. [Google Scholar] [CrossRef]
- Liang, J.; Cui, Y.; Wang, Q.; Geng, T.; Wang, W.; Liu, D. ClusterFomer: Clustering As A Universal Visual Learner. Adv. Neural Inf. Process. Syst. 2023, 36, 64029–64042. [Google Scholar]
- Caron, M.; Bojanowski, P.; Joulin, A.; Douze, M. Deep Clustering for Unsupervised Learning of Visual Features. In Proceedings of the Computer Vision—ECCV 2018: 15th European Conference, Proceedings, Part XIV, Munich, Germany, 8–14 September 2018; Springer: Berlin/Heidelberg, Germany, 2018; pp. 139–156. [Google Scholar]
- Ji, X.; Henriques, J.F.; Vedaldi, A. Invariant Information Clustering for Unsupervised Image Classification and Segmentation. arXiv 2018, arXiv:1807.06653. [Google Scholar]
- Guo, M.-H.; Cai, J.-X.; Liu, Z.-N.; Mu, T.-J.; Martin, R.R.; Hu, S.-M. PCT: Point cloud transformer. Comput. Vis. Media 2021, 7, 187–199. [Google Scholar] [CrossRef]
- Zheng, S.; Lu, J.; Zhao, H.; Zhu, X.; Luo, Z.; Wang, Y.; Fu, Y.; Feng, J.; Xiang, T.; Torr, P.H.S.; et al. Rethinking Semantic Segmentation from a Sequence-to-Sequence Perspective with Transformers. arXiv 2020, arXiv:2012.15840. [Google Scholar]
- de Bruijne, M.; Cattin, P.C.; Cotin, S.; Padoy, N.; Speidel, S.; Zheng, Y.; Essert, C. A Spatial Guided Self-supervised Clustering Network for Medical Image Segmentation. In Proceedings of theMedical Image Computing and Computer Assisted Intervention–MICCAI 2021: 24th International Conference, Strasbourg, France, 27 Septembe–1 October 2021; Springer International Publishing: Berlin/Heidelberg, Germany, 2021; pp. 379–388. [Google Scholar]
- Wang, Y.; Sun, Y.; Liu, Z.; Sarma, S.E.; Bronstein, M.M.; Solomon, J.M. Dynamic Graph CNN for Learning on Point Clouds. ACM Trans. Graph. 2019, 38, 1–12. [Google Scholar] [CrossRef]
- Xiang, T.; Zhang, C.; Song, Y.; Yu, J.; Cai, W. Walk in the Cloud: Learning Curves for Point Clouds Shape Analysis. In Proceedings of the IEEE/CVF International Conference on Computer Vision, Montreal, QC, Canada, 10–17 October 2021; pp. 915–924. [Google Scholar]
- Liu, Y.; Liang, K.; Xia, J.; Zhou, S.; Yang, X.; Liu, X.; Li, S.Z. Dink-Net: Neural Clustering on Large Graphs. In Proceedings of the 40th International Conference on Machine Learning, Proceedings of Machine Learning Research, Honolulu, HI, USA, 23–29 July 2023; pp. 21794–21812. [Google Scholar]
- Xu, M.; Zhang, J.; Peng, Z.; Xu, M.; Qi, X.; Qiao, Y. Learning Geometry-Disentangled Representation for Complementary Understanding of 3D Object Point Cloud. Proc. AAAI Conf. Artif. Intell. 2021, 35, 3056–3064. [Google Scholar] [CrossRef]
- Xue, Y.; Jin, G.; Shen, T.; Tan, L.; Wang, N.; Gao, J.; Wang, L. Consistent Representation Mining for Multi-Drone Single Object Tracking. IEEE Trans. Circuits Syst. Video Technol. 2024, 34, 10845–10859. [Google Scholar] [CrossRef]
- Kassem, H.; Hajjar, S.E.; Abdallah, F.; Omrani, H. Multi-view Deep Embedded Clustering: Exploring a new dimension of air pollution. Eng. Appl. Artif. Intell. 2025, 139, 18. [Google Scholar] [CrossRef]
- Bai, R.; Huang, R.; Xu, L.; Qin, Y. CSMDC: Exploring consistently context semantics for multi-view document clustering. Expert Syst. Appl. 2025, 261, 125386. [Google Scholar] [CrossRef]
- Xie, X.; Li, B.; Pan, E.; Guo, Z.; Kang, Z.; Chen, W. One Node One Model: Featuring the Missing-Half for Graph Clustering. In Proceedings of the AAAI Conference on Artificial IntelligenceArxiv, Pennsylvania, PA, USA, 25 February–4 March 2025; pp. 21688–21696. [Google Scholar]
- Chen, S.; Zhang, X.; He, Y.; Peng, L.; Ou, Y.; Xu, S. Multi-dimensional weighted deep subspace clustering with feature classification. Expert Syst. Appl. 2025, 260, 125375. [Google Scholar] [CrossRef]
- Gu, Q.; Niu, Y.; Hui, Z.; Wang, Q.; Xiong, N. A hierarchical clustering algorithm for addressing multi-modal multi-objective optimization problems. Expert Syst. Appl. 2025, 264, 125710. [Google Scholar] [CrossRef]
- Liu, Z.; Qiu, H.; Deveci, M.; Pedrycz, W.; Siarry, P. Multi-view neutrosophic c-means clustering algorithms. Expert Syst. Appl. 2025, 260, 125454. [Google Scholar] [CrossRef]
- Liu, Z.; Qiu, H.; Senapati, T.; Lin, M.; Abualigah, L.; Deveci, M. Enhancements of evidential c-means algorithms: A clustering framework via feature-weight learning. Expert Syst. Appl. 2025, 259, 125246. [Google Scholar] [CrossRef]
- Xie, X.; Chen, W.; Kang, Z.; Peng, C. Contrastive graph clustering with adaptive filter. Expert Syst. Appl. 2023, 219, 119645. [Google Scholar] [CrossRef]
- Dou, Z.; Peng, N.; Hou, W.; Xie, X.; Ma, X. Learning multi-level topology representation for multi-view clustering with deep non-negative matrix factorization. Neural Netw. 2025, 182, 106856. [Google Scholar] [CrossRef]
- Li, B.; Xie, X.; Lei, H.; Fang, R.; Kang, Z. Simplified PCNet with robustness. Neural Netw. 2025, 184, 107099. [Google Scholar] [CrossRef] [PubMed]
- Zhan, S.; Jiang, H.; Shen, D. Co-regularized optimal high-order graph embedding for multi-view clustering. Pattern Recognit. 2025, 157, 110892. [Google Scholar] [CrossRef]
- Zhu, Y.; Li, Q.; Liu, W.; Yin, C. Diffusion process with structural changes for subspace clustering. Pattern Recognit. 2025, 158, 111066. [Google Scholar] [CrossRef]
- Xie, X.; Chen, W.; Kang, Z. Robust graph structure learning under heterophily. Neural Netw. 2025, 185, 107206. [Google Scholar] [CrossRef]
- Wu, T.; Lu, G.-F. Specific and coupled double consistency multi-view subspace clustering with low-rank tensor learning. Signal Process. 2025, 229, 109803. [Google Scholar] [CrossRef]
- Cheng, S.; Chen, X.; He, X.; Liu, Z.; Bai, X. PRA-Net: Point Relation-Aware Network for 3D Point Cloud Analysis. IEEE Trans. Image Process. 2021, 30, 4436–4448. [Google Scholar] [CrossRef]
- Kaur, A.; Kumar, Y.; Sidhu, J. Exploring meta-heuristics for partitional clustering: Methods, metrics, datasets, and challenges. Artif. Intell. Rev. 2024, 57, 287. [Google Scholar] [CrossRef]
Parameter Name | Parameter Value |
---|---|
Iterations | 100 |
Batch size | 300 |
Optimizer | Adam |
Momentum | 0.9 |
Weight | 0.0001 |
Learning Rate | 0.001 |
Warm-up | 10 |
Centers | 100 |
Category | Models | Selection Rationale |
---|---|---|
Transformer-Based | Clusterformer [9], DeepCluster [10], IIC [11] | SOTA attention architectures for clustering |
Point Cloud Processing | PCT [12], CurveNet [16], GDANet [18] | Dominant methods for geometric data |
Graph-Based CV | DGCNN [15], SGSCN [14], Dink-Net [17] | Graph convolution benchmarks for structural data |
Multi-View Clustering | FPGC [22], SPCNet [29] | Top-cited multi-view frameworks (>300 citations) |
Architectural Innovations | SETR [13], PRANet [34] | Novel positional encoding and attention mechanisms |
Model | SSE | NMI |
---|---|---|
Clusterformer [9] | 1.4983 × 105 | 0.579 |
DeepCluster [10] | 1.5265 × 105 | 0.423 |
IIC [11] | 1.5065 × 105 | 0.382 |
PCT [12] | 1.5042 × 105 | 0.368 |
SETR [13] | 1.5025 × 105 | 0.582 |
SGSCN [14] | 1.5606 × 105 | 0.262 |
DGCNN [15] | 1.5241 × 105 | 0.423 |
CurveNet [16] | 1.5006 × 105 | 0.365 |
Dink-Net [17] | 1.4991 × 105 | 0.563 |
GDANet [18] | 1.4948 × 105 | 0.459 |
FPGC [22] | 1.4987 × 105 | 0.561 |
SPCNet [29] | 1.5012 × 105 | 0.447 |
PRANet [34] | 1.4978 × 105 | 0.517 |
Our method | 1.4790 × 105 | 0.593 |
Model | SSE | NMI |
---|---|---|
Clusterformer [9] | 2.1432 × 105 | 0.177 |
DeepCluster [10] | 2.2223 × 105 | 0.151 |
IIC [11] | 2.2143 × 105 | 0.161 |
PCT [12] | 2.1510 × 105 | 0.153 |
SETR [13] | 2.1345 × 105 | 0.175 |
SGSCN [14] | 2.3020 × 105 | 0.148 |
DGCNN [15] | 2.1871 × 105 | 0.145 |
CurveNet [16] | 2.4152 × 105 | 0.152 |
Dink-Net [17] | 2.2371 × 105 | 0.169 |
GDANet [18] | 2.1311 × 105 | 0.168 |
FPGC [22] | 2.1732 × 105 | 0.178 |
SPCNet [29] | 2.1985 × 105 | 0.160 |
PRANet [34] | 2.1433 × 105 | 0.179 |
Our method | 2.1283 × 105 | 0.185 |
Model | SSE | NMI |
---|---|---|
Clusterformer [9] | 1.0021 × 102 | 0.699 |
DeepCluster [10] | 1.0344 × 102 | 0.694 |
IIC [11] | 1.0213 × 102 | 0.625 |
PCT [12] | 9.7512 × 10 | 0.685 |
SETR [13] | 9.6769 × 10 | 0.723 |
SGSCN [14] | 9.7523 × 10 | 0.717 |
DGCNN [15] | 1.0122 × 102 | 0.701 |
CurveNet [16] | 9.7001 × 10 | 0.675 |
Dink-Net [17] | 1.0173 × 102 | 0.798 |
GDANet [18] | 9.6655 × 10 | 0.760 |
FPGC [22] | 1.0122 × 102 | 0.785 |
SPCNet [29] | 1.0039 × 102 | 0.710 |
PRANet [34] | 9.6523 × 10 | 0.735 |
Our method | 9.5436 × 10 | 0.853 |
Dataset | Models | SSE | NMI |
---|---|---|---|
Breast Cancer | CNN-only | 3.1526 × 105 | 0.297 |
No-FD | 2.5673 × 105 | 0.335 | |
No-FA | 2.8743 × 105 | 0.274 | |
No-AvgPool | 2.4542 × 105 | 0.368 | |
Our Method | 1.4790 × 105 | 0.593 | |
Heart Disease | CNN-only | 3.3203 × 105 | 0.157 |
No-FD | 2.7432 × 105 | 0.168 | |
No-FA | 2.9353 × 105 | 0.174 | |
No-AvgPool | 2.6742 × 105 | 0.181 | |
Our Method | 2.1283 × 105 | 0.185 | |
IRIS | CNN-only | 2.7103 × 105 | 0.377 |
No-FD | 2.1342 × 102 | 0.593 | |
No-FA | 2.3487 × 102 | 0.417 | |
No-AvgPool | 2.1043 × 102 | 0.601 | |
Our Method | 9.5436 × 10 | 0.853 |
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2025 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Shen, Z.; Jiao, Y.; Ji, A.; Ye, B.; Niu, Y.; Zuo, K.; Hu, P.; Li, W. A Novel Feature-Scheduling Aggregation Clustering Framework Based on Convolutional Neural Networks. Electronics 2025, 14, 2700. https://doi.org/10.3390/electronics14132700
Shen Z, Jiao Y, Ji A, Ye B, Niu Y, Zuo K, Hu P, Li W. A Novel Feature-Scheduling Aggregation Clustering Framework Based on Convolutional Neural Networks. Electronics. 2025; 14(13):2700. https://doi.org/10.3390/electronics14132700
Chicago/Turabian StyleShen, Zhangyi, Yu Jiao, Aohan Ji, Bingqing Ye, Yunfei Niu, Kaizhong Zuo, Peng Hu, and Wenjie Li. 2025. "A Novel Feature-Scheduling Aggregation Clustering Framework Based on Convolutional Neural Networks" Electronics 14, no. 13: 2700. https://doi.org/10.3390/electronics14132700
APA StyleShen, Z., Jiao, Y., Ji, A., Ye, B., Niu, Y., Zuo, K., Hu, P., & Li, W. (2025). A Novel Feature-Scheduling Aggregation Clustering Framework Based on Convolutional Neural Networks. Electronics, 14(13), 2700. https://doi.org/10.3390/electronics14132700