Publicaciones
Repositorio Institucional
Navegar
Slide 1 of 5 Publicación Acceso Abierto
Optimizing treatment to control LDL cholesterol using machine learning
(Computers in Biology and Medicine, 2025-06-16) Boneu Yepez, Deiby; Sierra Porta, David; Morales Aguas, Liz; Manzur Jattin, Fernando
Introduction: Increased LDL cholesterol is one of the main risk factors for cardiovascular diseases; therefore, adequate therapy reduces the risk of developing cardiovascular disease. Artificial intelligence (AI) is a tool that
can significantly help doctors select the optimal treatment aimed at each patient individually. Based on the above, the question arises: Which artificial intelligence model allows us to recommend the best treatment to
control LDL levels according to the patient’s cardiovascular risk factor? Methodology: The performance of various machine learning models was compared to assess their ability to predict the best-individualized therapy for each patient according to their cardiovascular risk from a registry of patients at a clinic specializing in cardiovascular diseases. The Machine learning models used included: RandomForestClassifier (RFC), GradientBoostClassifier (GBC), AdaBoostClassifier (ABC), ExtraTreeClassifier (ETC), Decision Tree Classifier (DTC), KNN, Support Vector Machine (SVM), Multinomial Logistic Regression (MLR), and Naive Bayes Classifier (NBC).
Population and sample: The records of 166 patients with any cardiovascular risk who had LDL alterations and used some therapeutic interventions were obtained. However, four medical records did not have creatinine levels;
therefore, they were excluded, counting at the end with 162 observations. Of these, a sample of 115 patients who achieved the LDL therapeutic goal was obtained. Results: The Random Forest Classifier (RFC) and Gradient Boosting Classifier (GBC) demonstrated superior performance in classifying optimal LDL-lowering therapy. In contrast, Naïve Bayes Classifier (NBC) overestimated outcomes and was deemed unsuitable. Conclusions: Machine learning models, particularly Random Forest Classifier, provide valuable tools for optimizing LDL control in high-risk cardiovascular patients. These models enhance clinical decision-making by enabling personalized therapy selection based on patient-specific risk factors.
Slide 2 of 5 Publicación Acceso Abierto
Derivative-aligned anticipation of forbush decreases from entropy and fractal markers
(Instrumentation and Methods for Astrophysics, 2026-01-14) Perez Navarro, Juan Diego; Sierra Porta, David
We develop a feature-based framework to anticipate Forbush decreases (FDs) in one-minute neutronmonitor records by tracking sliding-window invariants from information theory, scaling, and geometry. For each station we compute marker time series—including Shannon, spectral, approximate and sample entropy; Lempel–Ziv complexity; correlation dimension; and Higuchi and Katz fractal dimensions—smooth them with an exponentially weighted moving average, and analyze their within-station standardized first differences. Timing is referenced to an operational alignment time t0 defined as the minimum of the smoothed count first difference, and marker leads are reported in minutes (ℓ ∗ < 0 indicates anticipation). Station-level detectability is defined on a pre-t0 window using a robust z-score detector with bilateral threshold and persistence, requiring neither cross-correlation nor hypothesis testing. We apply the pipeline to two FD episodes with broad station coverage (2023-04-23 and 2024-05-10; 28 stations each). Across events, a compact CORE panel exhibits consistently high detection rates and predominantly negative lead distributions, with median leads of order several hours depending on the invariant and event. Lead dispersion across stations is substantial (interquartile ranges typically spanning a few hours), underscoring the value of station-wise criteria and distributional summaries rather than single-station inference. Representative marker trajectories confirm that early flagging corresponds to sustained pre-t0 excursions in marker differences, not merely tabulated artifacts. The approach is reproducible from open code, operates on native station units without cross-station
homogenization, and is qualitatively stable to sensitivity sweeps of windowing, smoothing, and detector parameters. These results support derivative-aligned invariant panels as practical early-warning complements to amplitude-threshold methods in space-weather nowcasting.
Subject headings: Forbush decrease, space weather, neutron monitor, sliding-window invariants, entropy measures, fractal dimension
Slide 3 of 5 Publicación Acceso Abierto
Complexity and scaling descriptors as diagnostic predictors of heliophysical indices across solar-cycle timescales
(Advances in Space Research, 2026-02-10) Sierra Porta, David; Canedo Verdugo, Maximiliano; Herrera Acevedo, Daniel David
Heliophysical variability emerges from a coupled, multiscale system in which changes in the solar atmosphere and heliospheric plasma translate into measurable signatures in widely used activity indices. Operational space-weather workflows often summarize this variability through
amplitudes and a small set of bulk solar-wind covariates, yet important dynamical information may also reside in the evolving morphology of the signals. We examine whether shape descriptors computed from heliophysical time series provide information beyond classical amplitude
summaries and standard bulk solar-wind covariates. Using daily OMNIWeb-era records spanning 1964–2025, we compute ten sliding-window descriptors under a past-only convention, designed to capture complementary aspects of temporal morphology such as irregularity, roughness, and long-range dependence. The descriptor set combines entropy measures, fractal-dimension estimators, the Hurst exponent, and Lempel–Ziv (LZ) complexity, yielding a compact representation of time-series structure that is not reducible to amplitude alone. The window length is treated as a methodological hyperparameter and selected through a target-specific sensitivity analysis that jointly favors competitive out-of-sample RMSE and stable permutation-importance rankings across neighboring windows. Two complementary learners, gradient boosting and a multilayer perceptron, are used as diagnostic probes to quantify permutation-based feature relevance under chronological splitting and training-only preprocessing. Across three targets (F10.7, Sunspot Number, and Dst), shape descriptors consistently rank among the most informative predictors, often matching or exceeding the relevance of standard solar-wind inputs. The most robust signals arise from LZ complexity and a compact subset of entropy/fractal measures, whose windowed trajectories track solar-cycle phases with characteristic lead–lag behaviour. Correlation analyses on both levels and standardised first differences expose redundancy within descriptor families and reduce spurious associations driven by shared nonstationarity, motivating a family-level interpretation of relevance rather than causal attribution. Overall, the results indicate that heliophysical time-series morphology encodes dynamical information complementary to amplitude- and bulk-plasma descriptions, suggesting compact, instrument-light features for augmenting future space-weather modelling pipelines.
Slide 4 of 5 Publicación Acceso Abierto
Urban mobility insights: A dataset for exploring network topology and city dynamics
(2025-09-11) Herrera Acevedo, Daniel; Sierra Porta, David
This article presents a comprehensive dataset capturing the urban network structures and sociodemographic variables of 65 cities worldwide for the year 2023, based on the Urban Mobility Readiness Index (UMRi) developed by the Oliver Wyman Forum. The dataset includes key metrics such as graph entropy, node degree, clustering coefficient, graph diameter, GDP per capita, and population density, among others, which are essential for analysing the relationship between network topology and urban mobility readiness. By offering detailed insights into these urban networks, this dataset serves as a valuable resource for cities not currently included in major mobility rankings, allowing them to evaluate their mobility readiness in relation to established indices like the UMRi. Urban planners and researchers can leverage this data to explore complex urban mobility dynamics and develop strategies to enhance transportation systems, particularly in rapidly growing or underserved regions. The dataset is structured for seamless integration with various analytical tools, making it a vital asset for both urban planning and research aimed at fostering sustainable and efficient urban development.
Slide 5 of 5 Publicación Acceso Abierto
Geomagnetic disturbances and grid vulnerability: Correlating storm intensity with power system failures
(2025-07-25) González Figueroa, Mauro A.; Herrera Acevedo, Daniel D.; Sierra Porta, David
Geomagnetic storms represent a critical yet sometimes overlooked factor affecting the reliability of modern power systems. This study examines the relationship between geomagnetic storm activity—characterized by the Dst index and categorized into weak, moderate, strong, severe, and extreme intensities—and reported power outages of unknown or unusual origin in the United States from 2006 to 2023. Outage data come from the DOE OE-417 Annual Summaries, while heliospheric and solar wind parameters (including proton density, plasma speed, and the interplanetary magnetic field) were obtained from NASA’s OMNIWeb database. Results indicate that years with a higher total count of geomagnetic storms, especially those featuring multiple strong or severe events, exhibit elevated incidences of unexplained power interruptions. Correlation analyses further reveal that increasingly negative Dst values, enhanced solar wind velocity, and higher alpha/proton ratios align with greater numbers of outages attributed to unknown causes, underscoring the pivotal role of solar wind–magnetosphere coupling. A simple regression model confirms that storm intensity and average magnetic field strength are statistically significant predictors of unexplained outages, more so than broad indicators such as sunspot number alone. These findings highlight the importance of monitoring high-intensity geomagnetic storms and associated heliospheric variables to mitigate potential risks. Greater attention to space weather impacts and improved reporting of outage causes could bolster grid resilience, helping operators anticipate and manage disruptions linked to geomagnetic disturbances.
