Marcelino Lázaro
Associate Professor
Signal Theory and Communications Department
University Carlos III of Madrid
Av. Universidad 30, 28911 Leganés - MADRID
SPAIN
Project PID2021-125652OB-I00
Grant PID2021-125652OB-I00 funded by:
Machine Learning for Smart Alarm Management (ML-SAM)
The ML-SAM project (Machine Learning for Smart Alarm Management) falls within the general framework of Machine Learning (ML). In particular, it encompasses all the research activities required to address a problem of current interest: the development of an intelligent alarm management system to assist operation centers responsible for large-scale critical infrastructures (such as those related to the generation, distribution, and supply of water, electricity, or gas; wide-area communication networks; transportation and logistics networks; emergency healthcare systems; or environmental monitoring).
Nowadays, network management is a business-oriented task, and operators need to find efficient strategies to rapidly resolve network incidents, thereby mitigating the impact of service degradation (direct economic impact, user satisfaction, etc.). With the deployment of new communication technologies (5G) and the expected increase in the number of connected elements, devices, and resources (especially with the advent of IoT), operators have focused their efforts on developing automated systems to support network management. These systems should include functionalities such as fault analysis (prediction, detection, and prioritization), alarm management (classification and clustering), and alert notification.
The specific characteristics of this problem require significant research efforts in several related areas. The underlying data are imbalanced, the environment is dynamic, and both binary classification and multiclass ordinal classification problems need to be addressed in such environments. The main objectives of the project are:
- To develop an alarm detection module, which involves the following technical objectives:
- To develop new theoretically grounded ML methods for the binary classification of imbalanced data
- To develop robust methods for dynamic scenarios affected by concept drift
- To integrate dynamic models with theoretically grounded ML classifiers
- To develop an intelligent alarm management system to support network operators in generating work orders for incident resolution, which requires:
- To develop new ML methods for multiclass and ordinal classification of imbalanced data
- To design new pattern clustering methods
There are also secondary research objectives related to the characteristics of the problem, which require data preprocessing and the use of Big Data Analytics, as well as objectives related to the requirements of the 'Right to Explanation' under EU regulations:
- To develop appropriate data preprocessing techniques for handling heterogeneous and incomplete data
- To implement the developed ML methods within a Big Data Analytics framework
- To develop methods for ML explainability
The project considers theoretically grounded methods for problems such as cost-sensitive classification and multiclass and ordinal classification, capable of operating in dynamic environments affected by concept drift and class imbalance. Therefore, these methods are applicable to numerous highly relevant problems in areas such as healthcare, economics, or any application requiring ratings or rankings. Advances in the explainability of learning methods will broaden the scope of ML applications in the context of new EU regulations.
Researchers
- Marcelino Lázaro (PI1)
- Francisco Javier González Serrano (PI2)
- Lorena Álvarez Pérez
- Harold Y. Molina Bulla
Project Details
- Funding Entities: Ministerio de Ciencia, Innovación y Universidades (MICIU)/Agencia Estatal de Investigación (AEI)/10.13039/501100011033, and the European Regional Development Fund (ERDF A way of making Europe)
- Reference: PID2021-125652OB-I00
- Execution: 01/09/2022 a 31/08/2026
Results
Below you can find some of the main results obtained during the project development.
Journal Publications
- Neural network for ordinal classification of imbalanced data by minimizing a Bayesian cost
Marcelino Lázaro, Aníbal R. Figueiras-Vidal
Pattern Recognition (2023)
https://doi.org/10.1016/j.patcog.2023.109303 - Optimum Bayesian thresholds for rebalanced classification problems using class-switching ensembles
Aitor Gutiérrez-López, Francisco Javier González-Serrano, Aníbal R. Figueiras-Vidal
Pattern Recognition (2023)
https://doi.org/10.1016/j.patcog.2022.109158 - One-step Bayesian example-dependent cost classification: The OsC-MLP method
Javier Mediavilla-Relaño, Marcelino Lázaro
Neural Networks (2024)
https://doi.org/10.1016/j.neunet.2024.106168 - COCOA: Cost-Optimized COunterfactuAl explanation method
Javier Mediavilla-Relaño, Marcelino Lázaro
Information Sciences (2024)
https://doi.org/10.1016/j.ins.2024.120616 - LSTM With Bayesian Loss for Ordinal and Imbalanced Channel Quality Prediction
Virginia Silva, Harold Molina-Bulla, Marcelino Lázaro, Mauricio Rodríguez, Matilde Sánchez-Fernández
IEEE Open Journan of the Communications Society (2025)
https://doi.org/10.1109/OJCOMS.2025.3619192 - Walsh sequences as direct Error Correcting Output Code dichotomies for multiclass problems
Lorena Álvarez-Pérez, Álvaro Callejas-Ramos
Engineering Applications of Artificial Intelligence (2026)
https://doi.org/10.1016/j.engappai.2026.114442 - A principled framework for multi-class imbalanced classification using asymmetric label switching and neutral rebalancing
Aitor Gutiérrez-López, Francisco Javier González-Serrano
Neural Computing and Applications (2026)
https://doi.org/10.1007/s00521-026-12395-3 - Context-Aware Alert Triage in Security Operations Centers: Escalation and Reporting Decisions under Label Inconsistency
Francisco Javier González-Serrano, Lorena Álvarez-Pérez, Marcelino Lázaro, Aitor Gutiérrez-López; José Luis Álvarez Aldana; Marta Gil-López; Andrés Izquierdo-Núñez; Jairo Montero-Santos
Enviado a Engineering Applications of Artificial Intelligence (En revisión) (2026)
Preprint SSRN: https://dx.doi.org/10.2139/ssrn.7536846
International Conferences
- Smart Incident Prediction from NOC Alert Events in Digital TV Broadcasting Networks
Francisco Javier González-Serrano, Lorena Álvarez-Pérez, Marcelino Lázaro, Aitor Gutiérrez-López; José Luis Álvarez Aldana; Marta Gil-López; Andrés Izquierdo-Núñez; Jairo Montero-Santos
Advances in Computational Intelligence. IWANN 2025. Springer LNCS-16009 (2025)
https://doi.org/10.1007/978-3-032-02728-3_33- Conference Presentation (PDF Slides)
- AI-Driven Alert Triage in Security Operations Centers: Imbalanced Learning with Human-in-the-Loop Contextual Bias Modeling
Marcelino Lázaro, Lorena Álvarez-Pérez, Francisco Javier González-Serrano, Aitor Gutiérrez-López; José Luis Álvarez Aldana; Marta Gil-López; Andrés Izquierdo-Núñez; Jairo Montero-Santos
Artificial Intelligence Applications and Innovations AIAI 2026. Springer IFIP Advances in Information and Communication Technology, vol 794 (2026)
https://doi.org/10.1007/978-3-032-30805-4_11- Conference Presentation (PDF Slides)
National (Spain) Conferences
- Sistema automático de predicción de incidencias en redes de difusión de TV digital
Lorena Álvarez-Pérez, Francisco Javier González-Serrano, Marcelino Lázaro, Harold Y. Molina-Bulla
URSI 2025. Simposio Nacional de la Unión Científica Internacional de Radio (2025)
- Conference Presentation (PDF Slides)
