IPIgnacio-Iker Prado-Rujas

Ignacio-Iker Prado-Rujas

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Ignacio-Iker Prado-Rujas has a total of 10 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Beyond the Error Metric: Enhancing Spatio-Temporal Sensor Data Forecasting with Non-Functional Features

    by Ignacio Iker Prado Rujas

    The presence of sensors that collect data is rapidly growing across various domains. They are used for a plethora of applications, particularly in the context of smart cities. This thesis focuses on fixed, geolocated time series sensor data. Such data can be leveraged to develop forecasting models that estimate future values observed by these sensors at a given time. This thesis studies the problem of spatio-temporal sensor data forecasting. The objective is to take advantage of multiple time series from neighboring sensors that measure the same variables. By doing so, forecasting models can benefit from a set of measurements across a region at each timestamp, rather than relying solely on the observations of a single sensor over time. Once the model is operational, its performance is typically evaluated using error metrics, such as the Root-Mean-Squared Error (RMSE). While such metrics are important, they neglect vital aspects of the model's usability and adaptability in real-world conditions. This thesis focuses on enhancing these characteristics, referred to as Non-Functional Features (NFFs). To tackle this problem, we propose a generic three-step methodology. First, we analyze and transform the geolocated time series data through seven predefined phases. The outcome of this step is a set of mesh-grids and relevant metadata. Next, we introduce the generic Spatio-Temporal Forecaster (ST-F) model, whose core component is a spatio-temporal module designed to extract features from the generated mesh-grids. Additionally, the ST-F model can incorporate exogenous modules to enhance its predictive capabilities, such as flat-extraction or recurrent ones. Finally, we devise how to evaluate the model using both conventional error metrics and the proposed NFFs. To validate our methodology, we apply it to three diverse spatio-temporal forecasting use cases. 1. Solar irradiance: We evaluate how the model adapts to a variable set of sensors and missing sensor observations. We denote these NFFs as flexibility and robustness. 2. Mobility demand: The second use case considers taxi and shared bicycle trips forecasting. Here, we introduce the extensibility NFF, which evaluates how the model benefits from combining different but related forecasted variables. Additionally, we study integrability, i.e., how the model improves the forecasts by incorporating exogenous data thanks to its modular structure. 3. Air quality: In the third use case, the ST-F model combines up to eleven pollutants, besides accomplishing additional NFFs. Overall, this thesis contributes to developing a methodology for understanding and forecasting spatio-temporal sensor data. Beyond achieving low error metrics, it seeks additional beneficial properties of forecasting models, enhancing their real-world applicability. RESUMEN La presencia de sensores que recogen datos está creciendo rápidamente en distintos dominios. Se utilizan para una gran variedad de aplicaciones, en particular en el contexto de las ciudades inteligentes. Esta tesis se centra en series temporales de datos de sensores fijos y geolocalizados. Estos datos pueden ser aprovechados para desarrollar modelos de predicción que estimen valores futuros observados por esos sensores en un momento determinado. Esta tesis estudia el problema de la predicción espacio-temporal de datos de sensores. El objetivo es aprovechar múltiples series temporales de sensores cercanos que midan las mismas variables. De este modo, los modelos de predicción se pueden beneficiar de un conjunto de medidas en toda una región en cada instante, en lugar de basarse únicamente en observaciones de un solo sensor a lo largo del tiempo. Una vez que el modelo está operativo, su rendimiento se evalúa típicamente mediante métricas de error, como la raíz del error cuadrático medio. Si bien dichas métricas son importantes, dejan de lado aspectos vitales como la usabilidad y adaptabilidad del modelo en condiciones reales. Esta tesis se centra en mejorar dichas características, referidas como atributos no funcionales. Para abordar este problema, proponemos una metodología genérica de tres pasos. Primero, analizamos y transformamos los datos geolocalizados de series temporales mediante siete fases predefinidas. El resultado de este paso es un conjunto de rejillas o matrices y metadatos relevantes. A continuación, presentamos el modelo genérico Spatio-Temporal Forecaster (ST-F), cuyo componente central es un módulo espacio-temporal que extrae características de las rejillas generadas. Adicionalmente, el modelo ST-F puede incorporar módulos exógenos para mejorar sus capacidades de predicción, como los de extracción plana o los recurrentes. Finalmente, diseñamos un método para evaluar el modelo utilizando tanto métricas de error convencionales como los atributos no funcionales propuestos. Para validar nuestra metodología, la aplicamos a tres casos de uso diversos sobre predicción espacio-temporal. 1. Irradiancia solar: Evaluamos cómo el modelo se adapta a un conjunto variable de sensores y a observaciones faltantes de los sensores. Denotamos esos atributos como flexibilidad y robustez. 2. Demanda de movilidad: El segundo caso de uso considera la predicción de viajes de taxi y bicicletas compartidas. Aquí, introducimos el atributo denominado extensibilidad, que evalúa cómo el modelo se beneficia de combinar distintas variables predichas relacionadas entre sí. Adicionalmente, estudiamos la integrabilidad, es decir, cómo el modelo mejora sus predicciones al incorporar datos exógenos gracias a su estructura modular. 3. Calidad del aire: En el tercer caso de uso, el modelo ST-F combina hasta once gases contaminantes, además de conseguir los otros atributos no funcionales. En conjunto, esta tesis contribuye a desarrollar...

  • Open Access

    A multivariable sensor-agnostic framework for spatio-temporal air quality forecasting based on Deep Learning

    by Ignacio-Iker Prado-Rujas, Antonio García-Dopico, Emilio Serrano, M. Luisa Córdoba, María S. Pérez


  • Open Access

    Combining heterogeneous data sources for spatio-temporal mobility demand forecasting

    by Ignacio-Iker Prado-Rujas, Emilio Serrano, Antonio García-Dopico, M. Luisa Córdoba, María S. Pérez


  • Open Access

    Mobility demand mesh-grid datasets

    by Ignacio-Iker Prado-Rujas

  • Open Access

    Mobility demand mesh-grid datasets

    by Ignacio-Iker Prado-Rujas

  • CHALLENGE-BASED LEARNING IN EXPLAINABLE ARTIFICIAL INTELLIGENCE EDUCATION

    by Ignacio-Iker Prado-Rujas, Emilio Serrano, Damiano Zanardini, Javier Bajo

    Information about the paper titled "CHALLENGE-BASED LEARNING IN EXPLAINABLE ARTIFICIAL INTELLIGENCE EDUCATION" at IATED Digital Library

  • Open Access

    An estimation of heavy-duty vehicle fleet CO2 emissions based on sampled data

    by Nikiforos Zacharof, Georgios Fontaras, Biagio Ciuffo, Alessandro Tansini, Iker Prado-Rujas


  • Open Access

    A Flexible and Robust Deep Learning-Based System for Solar Irradiance Forecasting

    by Ignacio-Iker Prado-Rujas, Antonio Garcia-Dopico, Emilio Serrano, Maria S. Perez

    Most studies about the solar forecasting topic do not analyze and exploit the temporal and spatial components that are inherent to such a task. Furthermore, they mostly focus just on precision and not on other meaningful features, such as flexibility and robustness. With the current energy production trends, where many solar panels are distributed across city rooftops, there is a need to manage all this information simultaneously and to be able to add and remove sensors as needed. Likewise, robust models need to be able to cope with (inevitable) sensor failure and continue producing reliable predictions. Due to all of this, solar forecasting models need to be as decoupled as possible from the number of data sources that feed them and their geographical distribution, enabling also the reusability of the models. This article contributes with a family of Deep Learning models for solar irradiance forecasting complying with the aforementioned features, i.e. flexibility and robustness. In the first stage, several Artificial Neural Networks are trained as a basis for predicting solar irradiance on several locations at the same time. Thereupon, a family of models that work with irradiance maps thanks to Convolutional Long Short-Term Memory layers is presented, obtaining forecast skills between 7.4% and 41% (depending on the location and horizon) compared to the baseline. The latter family comes with flexibility and robustness features, which are required in large-scale Intelligent Environments, such as Smart Cities. Working with irradiance maps means that new sensors can be added (or removed) as needed, without requiring rebuilding the model. Experiments carried out show that sensor failures have a mild impact on the prediction error for several forecast horizons.