Hybrid Deep Learning Ensemble with Dynamic Fusion for Reliable Anomaly Detection in Operational LTE RANs
Main Article Content
Keywords
Hybrid deep learning ensemble, LTE RAN KPI anomaly detection, Convolutional Autoencoder, Isolation Forest classification, SHAP interpretability
Abstract
Monitoring long-term evolution (LTE) network performance is increasingly complex due to rapidly growing data volumes and the diversity of quality-of-service indicators. Traditional monitoring approaches relying on static thresholds and manual key performance indicator (KPI) analysis often fail to detect multidimensional, evolving anomalies. We propose instead a hybrid deep ensemble learning framework for anomaly detection and diagnosis in Radio Access Networks (RANs). This framework integrates four complementary architectures: (i) a convolutional autoencoder (CAE); (ii) a Bidirectional Long Short-Term Memory AutoEncoder (BLSTM-AE); (iii) a transformer autoencoder (transformer AE); and (iv) a bidirectional LSTM forecaster, generating various anomaly scores. These scores are dynamically fused across frequency bands and processed with an Isolation Forest (IF) to produce the final anomaly judgment. An evaluation on real LTE data from Algerian mobile networks (three months, 1650 base stations, hourly KPIs) demonstrates the effectiveness of the proposed approach, achieving a maximum F1 score of 93.89%, an improvement of up to 9.5% over the best individual model. SHapley Additive exPlanations (SHAP)-based explainability analysis reveals that key operational indicators related to mobility and resource use drive the model’s decisions. This work provides a practical, interpretable hybrid framework validated on confidential operational data from a national operator under region-specific conditions.
References
Abrahams, T. O., Ewuga, S. K., Kaggwa, S., Uwaoma, P. U., Hassan, A. O., & Dawodu, S. O. (2023). Review of strategic alignment: Accounting and cybersecurity for data confidentiality and financial security. World Journal of Advanced Research and Reviews, 20(03), 1743–1756. https://doi.org/10.30574/wjarr.2023.20.3.2691.
Ahmed, M. J., Mozó, A., & Karamchandani, A. (2025). A survey on graph neural networks, machine learning and deep learning techniques for time series applications in industry. PeerJ, 11, e3097. https://doi.org/10.7717/peerj-cs.3097.
Bordeau-Aubert, K., Whatley, J., Nadeau, S., Glatard, T., & Jaumard, B. (2023). Classification of anomalies in telecommunication network KPI time series. arXiv preprint arXiv:2308.16279. https://doi.org/10.48550/arXiv.2308.16279
Chen, J., Zhang, J., Qian, R., Yuan, J., & Ren, Y. (2023). An Anomaly Detection Method for Wireless Sensor Networks Based on the Improved Isolation Forest. Applied Sciences, 13(2), 702. https://doi.org/10.3390/app13020702
Ericsson. (n.d.). Ericsson Network Manager. Accessed 31 May 2025. Available at https://portfolio.ericsson.net/c/FGB1010381
ETSI. (2022, April). TS 132 450 V17.0.0 — Universal Mobile Telecommunications System (UMTS); LTE; Telecommunication management; Key Performance Indicators (KPI) for Evolved Universal Terrestrial Radio Access Network (E-UTRAN): Definitions (3GPP TS 32.450 version 17.0.0 Release 17). https://www.etsi.org/deliver/etsi_ts /132400_132499/132450/17.00.00_60/ts_132450v170000p.pdf
Fan, J., Liu, Z., Wu, H., Wu, J., Si, Z., Hao, P., & Luan, T. H. (2023). Luad: A lightweight unsupervised anomaly detection scheme for multivariate time series data. Neurocomputing,557, 126644. https://doi.org/10.1016/j.neucom.2023.126644
Gao, F., Xin, R., & Zhang, Y. (2025). Are GNNs Effective for Multimodal Fault Diagnosis in Microservice Systems? arXiv preprint arXiv:2501.02766 https://doi.org/10.48550/arxiv.2501.02766
Hasan, A., Boeira, C., Papry, K., Ju, Y., Zhu, Z., & Haque, I. (2024). Root cause analysis of anomalies in 5g ran using graph neural network and transformer. arXiv preprint arXiv:2406.15638. https://arxiv.org/abs/2406.15638
Imoize, A. L., Orolu, K., & Atayero, A. A.-A. (2020). Analysis of key performance indicators of a 4G LTE network based on experimental data obtained from a densely populated smart city. Data in Brief, 29. https://doi.org/10.1016/j.dib.2020.105304
Isaac, E. R., & Sharma, A. (2024). Adaptive thresholding heuristic for KPI anomaly detection. In 2024 16th International Conference on Communication Systems Networks (COMSNETS) (pp. 737–741). https://doi.org/10.1109/COMSNETS59351.2024.10427016
Isaac, E. R., & Singh, B. (2025, January). QBSD: Quartile-Based Seasonality Decomposition for Cost-Effective RAN KPI Forecasting. In 2025 17th International Conference on COMmunication Systems and NETworks (COMSNETS) (pp. 847–851). IEEE. https://doi.org/10.1109/COMSNETS63942.2025.10885747
Jiang, H., Ji, S., He, G., & Li, X. (2023). Network traffic anomaly detection model based on feature reduction and bidirectional LSTM neural network optimization. Scientific Programming, 2023(1), 2989533. https://doi.org/10.1155/2023/2989533
Ma, L., Dong, J., Peng, K., & Zhang, C. (2019). Hierarchical Monitoring and Root-Cause Diagnosis Framework for Key Performance Indicator-Related Multiple Faults in Process Industries. IEEE Transactions on Industrial Informatics, 15(4), 2091–2100. https://doi.org/10.1109/TII.2018.2855189.
Oshin, O., Luka, M. K., & Atayero, A. A. (2016). From 3GPP LTE to 5G: An Evolution. In: Ao, Si., Yang, G. C., & Gelman, L. (eds), Transactions on Engineering Technologies. (pp. 485–502). Springer, Singapore. https://doi.org/10.1007/978-981-10-1088-0_36
Saidi, F. A., Phinzi, K., & Molnár, E. (2023). Urbanization in Algeria: Toward a More Balanced and Sustainable Urban Network? Social Sciences, 12(3), 174. https://doi.org/10.3390/socsci12030174
Scatà, M. (2025). Complex Systems in Communication Networks. In The Power of Complex Systems: Unlocking Intelligence in Mobile Communication Networks (pp. 1–33). Cham: Springer Nature Switzerland. https://doi.org/10.1007/978-3-032-04243-9_1
Stoklasa, R., Calabrese, N., & Serio, L. (2024). Investigation and perspectives of using Graph Neural Networks to model complex systems: the simulation of the helium II bayonet heat exchanger in the LHC. IOP Conference Series: Materials Science and Engineering, 1301, 012157. https://doi.org/10.1088/1757-899x/1301/1/012157
Sun, N., Zheng, H., Chen, Y., Liu, Y., & Fang, J. (2023). Accurate Anomaly Interval Recognition and Fault Classification by Pattern Mining and Clustering. IEEE INFOCOM 2023 — IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS), pp. 1–6. https://doi.org/10.1109/infocomwkshps57453.2023.10226079
Verhaelen, B., Mayer, F., Peukert, S., & Lanza, G. (2021). A comprehensive KPI network for the performance measurement and management in global production networks. Production Engineering, 15, – 650. https://doi.org/10.1007/s11740-021-01041-7
Youden, W. J. (1950). Index for rating diagnostic tests. Cancer, 3(1), 32–35.
Zhao, N., Zhu, J., Liu, R., Liu, D., Zhang, M., & Pei, D. (2019). Label-Less: A Semi-Automatic Labelling Tool for KPI Anomalies. International Conference on Computer Communications, pp. 1882–1890. https://doi.org/10.1109/INFOCOM.2019.8737429
Zou, C., Yuan, A., & Hu, J. (2024, September). BiLSTM-Based Anomaly Detection in Multivariate Time Series with Attention Mechanism and Dual Analysis. In 2024 IEEE 7th International Conference on Information Systems and Computer Aided Education (ICISCAE) (pp. 379–384). IEEE. https://doi.org/10.1109/ICISCAE62304.2024.10761506
Article Sidebar
Article Details
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Copyright Telecommunications Association Inc.
Insaf Boumezbeur, Centre de Recherche en Aménagement du Territoire (CRAT), Campus Zouaghi Slimane, Constantine, Algerie
Centre de Recherche en Aménagement du Territoire (CRAT), Campus Zouaghi Slimane, Route de Ain el Bey, 25000 Constantine, Algerie
How to Cite
