Pasar al contenido principal

FORTALECIENDO LA IA CON MEJORES DATOS Y MODELOS MAS CONFIABLES

PID2023-148396NB-I00

•
Nombre agencia financiadora Agencia Estatal de Investigación
Acrónimo agencia financiadora AEI
Programa Programa Estatal para Impulsar la Investigación Científico-Técnica y su Transferencia
Subprograma Subprograma Estatal de Generación de Conocimiento
Convocatoria Proyectos de I+D+i (Generación de Conocimiento y Retos Investigación)
Año convocatoria 2023
Plan Estatal 2021-2023 Plan Estatal de Investigación Científica y Técnica y de Innovación
Centro beneficiario UNIVERSIDAD DE CORDOBA

Resultados relacionados

Resultados totales (Incluyendo duplicados): 9
Encontrada(s) 1 página(s)

Helvia. Repositorio Institucional de la Universidad de Córdoba
Publicación Artículo científico (JournalArticle)

Towards accessible AI for addressing students’ academic dropout: an auto machine learning and explainable artificial intelligence approach

  • Buñay-Guisñan, Pamela
  • Lara, Juan A.
  • Cano, Alberto
  • Cerezo, Rebeca
  • Romero, Cristóbal
Embargado hasta 02/01/2027, Predicting students at risk of dropping out and intervening on them has become an important challenge, in which Artificial Intelligence techniques have shown promise. However, many teachers and other stakeholders are not experts in those techniques and face notable technical obstacles accessing them. To make Ar-tificial Intelligence more accessible for the prediction of and the intervention on at-risk students, we propose the integration of Automated Machine Learning and Ex-plainable Artificial Intelligence techniques into a Visual Interactive Dashboard to automate the entire process. Our objective is to allow non-expert users to easily obtain predictions from data that will help make decisions to avoid student dropout. Our dashboard tailored the interface providing two views (denoted basic and ad-vanced) intended to be used for beginners and intermediate users respectively. In this paper we describe a case study in which a group of 49 users ran the Dashboard with a public dataset with the assignment to take action about student dropout. The Dashboard demonstrated promising performance. All the advanced interface users and 87.5% of the basic interface users completed the required task successfully in 4 to 6 minutes. A system usability evaluation produced scores of 78.7 and 74.5 out of 100 for the advanced and basic interfaces respectively. Finally, there were no significant differences between the two groups in terms of task success, time on task, or usability. These results demonstrate a contribution of the presented ap-proach in the accessibility process of non-expert users for addressing the problem of student’s academic dropout.
Helvia. Repositorio Institucional de la Universidad de Córdoba
Publicación Artículo científico (JournalArticle)

Group Counterfactual Explanations: A Use Case to Support Students at Risk of Dropping Out in Online Education

  • Buñay-Guisñan, Pamela
  • Cano, Alberto
  • Anguera, Aurea
  • Lara, Juan A.
  • Romero, Cristóbal
This paper proposes the novel application of group counterfactual explanations to the problem of predicting students at risk of dropout. Our objective is to explain how to recover the largest possible number of students while minimizing effort and cost. Using group counterfactuals, instructors and institutions could recover large groups of students with minimal remedial actions. For testing, we used the well-known public educational Open University Learning Analytics Dataset (OULAD), which contains students’ clicks made throughout interactions with online courses. We modified and adapted the only existing algorithm for the generation of group counterfactuals, named GROUP-CF. We also used the Diverse Counterfactual Explanations (DiCE) individual counterfactual algorithm with the K-means clustering method and new options in discovering the most representative counterfactuals for a group of students. The results obtained are very promising; our approach can be successfully applied to recover 99.3% of students at risk of failing in a shorter time in comparison to traditional individual counterfactuals. Moreover, although a group counterfactual proposes to change a greater number of students’ features, the values are lighter and therefore seem easier to apply than the ones obtained with individual counterfactuals. This work opens up a new line of research in education.
Helvia. Repositorio Institucional de la Universidad de Córdoba
Publicación Artículo científico (JournalArticle)

Counterfactual explanations in Education: a systematic review

  • Buñay-Guisñan, Pamela
  • Lara, Juan A.
  • Romero, Cristóbal
Counterfactuals are a type of explanations based on hypothetical scenarios used in Explainable Artificial Intelligence (XAI), showing what changes in input variables could have led to different outcomes in predictive problems. In the field of education, counterfactuals enable educators to explore various hypothetical scenarios, facilitating informed decision-making and the application of educational strategies for improving students' academic performance or reducing dropout rates, among others. Despite the gradual expansion of research on counterfactuals in education, systematic literature reviews on this topic remain scarce. The identification of the most relevant advancements in this field can provide a deep insight into the current state of research, highlighting the most effective areas and revealing opportunities for future studies. The objective of this research is to conduct a systematic literature review, using the PRISMA methodology, to analyse three aspects regarding the use of counterfactuals in education: the problems that counterfactuals help to address in education, the methods and/or algorithms used to generate them, and how the counterfactuals are presented in the educational context. As a result, we have identified a series of key challenges and opportunities for future research over the next few years, which constitutes the main contribution of this paper.
Helvia. Repositorio Institucional de la Universidad de Córdoba
Publicación Artículo científico (JournalArticle)

Dropout insight: Educational risk dashboard with counterfactual explanations

  • Muñoz-Muñoz, Marta
  • Luna, Christian
  • Lara, Juan A.
  • Romero, Cristóbal
The prediction and prevention of students at risk of dropout are two of the most important challenges in the educational domain. Although some commercial predictive tools support at-risk estimation and provide explanations of the associated factors, none of them offer recommendations to address or reverse potential dropout cases. This paper proposes Dropout Insight as a prescriptive web-based interactive tool that automates the entire data-mining process to suggest specific decisions. It supports the loading and processing of student data, the selection of the best predictive model, and the visualization of results through interpretation techniques based on explainers. The tool provides a clear and visually intuitive interface that enables users to explore risk factors and simulate alternative scenarios, including instructors and other stakeholders, without prior knowledge of data mining. It offers not only traditional individual counterfactual explanations, but also novel group counterfactuals, which generate hypothetical clusters or groups of students with similar behavioral profiles. These groups help recover the largest possible number of at-risk students with less effort and cost by offering a single, shared recommendation for intervention. By integrating automated prediction tools with visual, explainable artificial intelligence methods and counterfactual reasoning, the tool becomes a highly valuable and innovative resource to support pedagogical decision-making and guide proactive educational policies aimed at preventing dropout.
Helvia. Repositorio Institucional de la Universidad de Córdoba
Publicación Artículo científico (JournalArticle)

A two-stage framework for early failure detection in predictive maintenance: A case study on metro trains

  • Toribio, Luis
  • Veloso, Bruno
  • Gama, João
  • Zafra Gómez, Amelia
Early fault detection remains a critical challenge in predictive maintenance (PdM), particularly within critical infrastructure, where undetected failures or delayed interventions can compromise safety and disrupt operations. Traditional anomaly detection methods are typically reactive, relying on real-time sensor data to identify deviations as they occur. This reactive nature often provides insufficient lead time for effective maintenance planning. To address this limitation, we propose a novel two-stage early detection framework that integrates time series forecasting with anomaly detection to anticipate equipment failures several hours in advance. In the first stage, future sensor signal values are predicted using forecasting models; in the second, conventional anomaly detection algorithms are applied directly to the forecasted data. By shifting from real-time to anticipatory detection, the framework aims to deliver actionable early warnings, enabling timely and preventive maintenance. We validate this approach through a case study focused on metro train systems, an environment where early fault detection is crucial for minimizing service disruptions, optimizing maintenance schedules, and ensuring passenger safety. The framework is evaluated across three forecast horizons (1, 3, and 6 hours ahead) using twelve state-of-the-art anomaly detection algorithms from diverse methodological families. Detection performance is assessed using five performance metrics. Results show that anomaly detection remains highly effective at short to medium horizons, with performance at 1-hour and 3-hour forecasts comparable to that of real-time data. Ensemble and deep learning models exhibit strong robustness to forecast uncertainty, maintaining consistent results with real-time data even at 6-hour forecasts. In contrast, distance- and density-based models suffer substantial degradation at longer horizons (6-hours), reflecting their sensitivity to distributional shifts in predicted signals. Overall, the proposed framework offers a practical and extensible solution for enhancing traditional PdM systems with proactive capabilities. By enabling early anomaly detection on forecasted data, it supports improved decision-making, operational resilience, and maintenance planning in industrial environments.
Helvia. Repositorio Institucional de la Universidad de Córdoba
Publicación Artículo científico (JournalArticle)

Deep learning for multivariate time series anomaly detection: an evaluation of reconstruction-based methods

  • Yahya, Mohammed Amer
  • Moya Martín-Castaño, Antonio Rafael
  • Ventura Soto, S.
In the field of anomaly detection in time series, remarkable advances based on deep learning methodologies and, more specifically, reconstruction-based methods have been proposed. These methods are particularly valuable, as they can capture the fundamental structure of the data and enable the detection of subtle anomalies that traditional techniques might overlook. Reviews of the existing literature discuss anomaly detection from a general perspective and consider a specific anomaly type, hindering the process of proposing new and better algorithms. This paper focuses on reconstruction-based methods in isolation, as they have been demonstrated to present the best performance of the three main groups in deep anomaly detection models described so far. Thus, it intends to extend the literature by addressing in detail reconstruction-based methods for anomaly detection in multivariate time series, to provide richer information about these methods, and to include extensive experimentation, not usually performed in existing surveys on the topic. Finally, the paper presents useful insights (strengths and weaknesses of the methods) extracted from the experimental study, and significant challenges and future research directions related to the potential of these methods in anomaly detection.
Helvia. Repositorio Institucional de la Universidad de Córdoba
Publicación Artículo científico (JournalArticle)

Comparative of machine learning methods for detecting cardiovascular disease

  • Ávila Jiménez, José Luis
  • Rodriguez-Lozano, Francisco J.
  • Cantón-Habas, Vanesa
  • Ortiz López, Manuel
Datos de investigación disponibles en: GitHub repository: https://github.com/FJ-Rodriguez-Lozano/Comparative-Machine-Learning-Cardiovascular-Disease, Background : Cardiovascular disease (CVD) is a leading cause of death globally, making its early diagnosis and classification critical in healthcare management due to its high prevalence. Objective: The goal of the study is to explore the effectiveness of various machine learning algorithms in the early diagnosis and classification of cardiovascular disease, focusing on their performance and interpretability, to be used as an accurate screening model for CVD. Methods: A stratified cross-validation methodology has been employed to assess the performance of several machine learning algorithms rigorously. The analysis included both simpler models and more complex decision tree-based algorithms. Results: The study revealed significant performance disparities among the algorithms. Simpler models like Naive Bayes and One Rule approached an acceptable 90% accuracy threshold. However, decision tree-based algorithms, particularly Logistic Models Trees (LMT), demonstrated the highest predictive performance and strong stability across different data subsets. LMT, integrating decision trees with logistic regressions, achieved the highest predictive performance among the evaluated algorithms. Conclusions: The high predictive performance and interpretability of LMT suggest that it may be a promising model for supporting clinical decision-making in CVD screening. The study advocates for the strategic selection of decision tree-based algorithms to enhance diagnostic precision and patient outcomes in CVD. Highlighting the superior performance and interpretability of these models underlines the importance of thoughtful algorithm selection in the fight against one of the foremost global health challenges.
Helvia. Repositorio Institucional de la Universidad de Córdoba
Publicación Artículo científico (JournalArticle)

Automated machine learning for test case prioritisation

  • Romero, José Raúl
  • Ramírez, Aurora
  • García-Martínez, Carlos
Test case prioritisation (TCP) involves ordering and selecting the most relevant test cases to verify that the current functionality of a software system remains unaffected by code changes. Recently, TCP has been addressed by machine learning (ML), predicting the failure probability of each test case. However, software engineers may struggle to identify and implement the most suitable predictive models for TCP. As new builds adapt the test suite being tested, the model performance may decline with the incorporation of these new builds. In this study, we address these challenges by applying automated workflow composition, including algorithm selection and hyperparameter optimisation. They are considered tasks within automated machine learning (AutoML). With this aim, our proposal employs grammar-guided genetic programming as the underlying mechanism for implementing the AutoML algorithm. Our experimental results demonstrate that our approach can adapt to the particularities of the system under test, selecting the most appropriate ML pipeline and hyperparameters for each build. More importantly, our approach reduces the ML knowledge required by testers---particularly regarding the selection and tuning of algorithms---while enabling them to generate pipelines suited to successive changes in SUT builds. This research showcases the potential of AutoML in software engineering, specifically for the TCP problem.
Helvia. Repositorio Institucional de la Universidad de Córdoba
Publicación Tesis doctoral (DoctoralThesis)

New deep learning approaches in anomaly detection. Applications, Nuevos enfoques de aprendizaje profundo en la deteccion de anomalias. Aplicaciones

  • Yahya, Mohammed Amer
This thesis investigates new deep learning approaches for anomaly detection, with a particular focus on multivariate time series data and reconstruction-based unsupervised methods. Anomaly detection in time series is a critical task in many real-world applications, including industrial monitoring, cybersecurity, healthcare, and large-scale information systems, where abnormal behaviors may indicate faults, attacks, or system degradation. The problem is especially challenging due to the rarity and diversity of anomalies, the high dimensionality of modern data, the presence of noise and non-stationarity, and the limited availability of labeled data. The thesis addresses these challenges through a combination of methodological rigor and model design innovation. First, it provides a comprehensive and unified experimental study of reconstruction-based deep learning methods for time series anomaly detection. A wide range of representative models—covering Autoencoders (AE), Variational Autoencoders (VAE), GAN-based reconstructors, and Transformer-based architectures—are evaluated under a common experimental protocol. All models are tested on heterogeneous multivariate datasets using consistent preprocessing, windowing, and evaluation criteria. To ensure scientific robustness, the comparison is supported by non-parametric statistical tests (Friedman and Nemenyi), allowing performance differences to be assessed beyond anecdotal results. This study clarifies the strengths, weaknesses, and typical failure modes of different reconstruction paradigms across diverse scenarios. Building on the insights obtained from this unified evaluation, the thesis proposes a new Two-Phase anomaly detection framework based on a time–frequency Transformer architecture. The proposed approach combines temporal and spectral representations of time series while preserving timestamp-level anomaly localization. A key contribution is the introduction of a diagnostics-driven design philosophy, in which additional model components are activated only when justified by measurable data characteristics. Specifically, the framework includes three optional modules: (i) an inter-metric embedding mechanism to capture cross-variable dependencies, (ii) a contamination-aware pseudo-filtering strategy to mitigate the effect of anomalous samples within nominal training data, and (iii) an attention-based fusion mechanism to adaptively combine time-domain and frequency-domain anomaly scores. The Two-Phase framework separates diagnostic analysis from model training and inference, reducing the risk of unnecessary model complexity and improving robustness across datasets with different statistical properties. Extensive experiments demonstrate that the proposed method achieves strong average performance across multiple benchmarks and obtains favorable statistical rankings compared to established baselines. At the same time, the thesis explicitly discusses the limitations of reconstruction-based approaches, including sensitivity to threshold selection, training data contamination, and challenging datasets with complex long-range dependencies. Overall, this thesis contributes both a rigorous empirical foundation for understanding reconstruction-based deep anomaly detection methods and a practical, data-aware modeling framework that improves robustness and interpretability. The results provide valuable guidance for researchers and practitioners working on anomaly detection in complex multivariate time series, and they open several directions for future research, including streaming settings, concept drift adaptation, uncertainty estimation, and interpretability of anomaly scores., Esta tesis se centra en el estudio y desarrollo de nuevos enfoques de aprendizaje profundo para la detección de anomalías, con especial atención a series temporales multivariantes y a métodos no supervisados basados en reconstrucción. La detección de anomalías en series temporales es una tarea clave en numerosos ámbitos del mundo real, como la monitorización industrial, los sistemas ciberfísicos, la ciberseguridad, la sanidad y los sistemas de información a gran escala. En estos contextos, las anomalías pueden estar asociadas a fallos, ataques o comportamientos inesperados, por lo que su detección temprana resulta esencial para garantizar la fiabilidad y seguridad de los sistemas. Este problema presenta importantes desafíos metodológicos. Las anomalías suelen ser raras y heterogéneas, los datos son habitualmente de alta dimensionalidad y los sistemas reales están sujetos a ruido y cambios de distribución a lo largo del tiempo. Además, en muchos escenarios prácticos no se dispone de datos etiquetados o estos son muy limitados, lo que dificulta la aplicación de enfoques supervisados. Por estas razones, los métodos no supervisados basados en reconstrucción se han convertido en una alternativa ampliamente utilizada, aunque su evaluación comparativa sigue siendo compleja debido a la falta de protocolos experimentales homogéneos. La primera contribución principal de esta tesis consiste en un estudio experimental exhaustivo y sistemático de métodos de detección de anomalías basados en reconstrucción mediante aprendizaje profundo. Se analizan modelos representativos de distintas familias, incluyendo Autoencoders (AE), Autoencoders Variacionales (VAE), modelos basados en redes generativas adversarias (GAN) y arquitecturas basadas en Transformers. Todos los métodos se evalúan bajo un protocolo experimental común, que incluye procedimientos consistentes de preprocesamiento, segmentación mediante ventanas temporales, entrenamiento y evaluación. El análisis se realiza sobre un conjunto diverso de series temporales multivariantes, que difieren en número de variables, dominio de aplicación y proporción de anomalías. Para garantizar la solidez científica de las conclusiones, la comparación se apoya en pruebas estadísticas no paramétricas, lo que permite evaluar de manera rigurosa la significación de las diferencias observadas entre los modelos. A partir de las conclusiones extraídas de este estudio comparativo, la tesis propone una nueva estrategia de detección de anomalías en dos fases, orientada a mejorar la robustez y adaptabilidad de los enfoques basados en reconstrucción. El método se fundamenta en una arquitectura Transformer tiempo–frecuencia, que combina información del dominio temporal y del dominio frecuencial, manteniendo al mismo tiempo una localización precisa de las anomalías a nivel de instante temporal. Esta combinación resulta especialmente adecuada para la detección de anomalías colectivas y patrones anómalos complejos. Una aportación clave del enfoque propuesto es la introducción de una fase de diagnóstico previa al entrenamiento, en la que se analizan propiedades estadísticas de los datos con el fin de decidir si deben activarse determinados módulos adicionales del modelo. Estos módulos están diseñados para capturar dependencias entre variables, mitigar el impacto de datos contaminados durante el entrenamiento y combinar de forma adaptativa las puntuaciones de anomalía obtenidas en los dominios temporal y frecuencial. Esta filosofía de diseño permite evitar la introducción innecesaria de complejidad y reduce el riesgo de degradación del rendimiento en conjuntos de datos que no se benefician de dichas extensiones. Los resultados experimentales muestran que el enfoque propuesto alcanza un rendimiento competitivo y consistente en múltiples conjuntos de datos de referencia, obteniendo clasificaciones estadísticas favorables frente a métodos consolidados. Al mismo tiempo, la tesis analiza de forma explícita las limitaciones inherentes a los enfoques basados en reconstrucción, como la sensibilidad a la selección de umbrales y la contaminación de los datos de entrenamiento.