Search

Ambient air pollution is a pervasive global issue that poses significant health risks. Among pollutants, ozone (O3) is responsible for an estimated 1 to 1.2 million premature deaths yearly. Furthermore, O3 adversely affects climate warming, crop productivity, and more. Its formation occurs when nitrogen oxides and volatile organic compounds react with short-wavelength solar radiation. Consequently, urban areas with high traffic volume and elevated temperatures are particularly prone to elevated O3 levels, which pose a significant health risk to their inhabitants. In response to this problem, many countries have developed web and mobile applications that provide real-time air pollution information using sensor data. However, while these applications offer valuable insight into current pollution levels, predicting future pollutant behavior is crucial for effective planning and mitigation strategies. Therefore, our main objectives are to develop accurate and efficient prediction models and identify the key factors that influence O3 levels. We adopt a time series forecasting approach to address these objectives, which allows us to analyze and predict future O3 behavior. Additionally, we tackle the feature selection problem to identify the most relevant features and periods that contribute to prediction accuracy by introducing a novel method called the Time Selection Layer in Deep Learning models, which significantly improves model performance, reduces complexity, and enhances interpretability. Our study focuses on data collected from five representative areas in Seville, Cordova, and Jaen provinces in Spain, using multiple sensors to capture comprehensive pollution data. We compare the performance of three models: Lasso, Decision Tree, and Deep Learning with and without incorporating the Time Selection Layer. Our results demonstrate that including the Time Selection Layer significantly enhances the effectiveness and interpretability of Deep Learning models, achieving an

Ensuring the optimal performance of power transformers is a laborious task in which the insulation system plays a vital role in decreasing their deterioration. The insulation system uses insulating oil to control temperature, as high temperatures can reduce the lifetime of the transformers and lead to expensive maintenance. Deep learning architectures have been demonstrated remarkable results in various fields. However, this improvement often comes at the cost of increased computing resources, which, in turn, increases the carbon footprint and hinders the optimization of architectures. In this study, we introduce a novel deep learning architecture that achieves a comparable efficacy to the best existing architectures in transformer oil temperature forecasting while improving efficiency. Effective forecasting can help prevent high temperatures and monitor the future condition of power transformers, thereby reducing unnecessary waste. To balance the inductive bias in our architecture, we propose the Smooth Residual Block, which divides the original problem into multiple subproblems to obtain different representations of the time series, collaboratively achieving the final forecasting. We applied our architecture to the Electricity Transformer datasets, which obtain transformer insulating oil temperature measures from two transformers in China. The results showed a 13% improvement in MSE and a 57% improvement in performance compared to the best current architectures, to the best of our knowledge. Moreover, we analyzed the architecture behavior to gain an intuitive understanding of the achieved solution.

This thesis focuses on the study of time series prediction using the technique known as deep learning or neural networks. At the same time, a series of new methodological proposals are made, which improve the efficiency of existing architectures, applied to a series of real data sets that present a challenge today. The technique known as deep learning has gained great popularity in recent years due to its incredible results in areas such as computer vision, natural language processing and time series prediction, among others. This technique is inspired by the functioning of the basic brain cell, the neuron. Neurons are organized in layers forming a neural network, processing the input information and propagating its output to other layers of neurons until the final output is obtained. This technique has been adapted on multiple occasions to the prediction of time series, developing architectures with results that are competitive with the current state of the art. However, although effectiveness has been a great advantage, sometimes these architectures have degraded their efficiency, preventing their application in real scenarios. There are several ways to improve efficiency, reducing some of the aspects that take a large number of resources such as: memory needed to store the architecture, inference time or training time, among others. This thesis focuses on improving training time, since it is the bottleneck when experimenting with new architectures, optimizing existing architectures, or retraining architectures in certain real scenarios. Faced with the problem of efficiency presented by architectures in the field of deep learning or neural networks, four different proposals have been made, whose main objective is to obtain greater efficiency by obtaining equal or superior effectiveness with respect to the architectures used in the comparative analysis. The first of the proposals introduces the idea of incremental learning into the design of the architecture. This, La presente tesis se centra en el estudio de la predicción de series temporales mediante el uso de la técnica conocida como deep learning (aprendizaje profundo en español) o redes neuronales. A su vez, se realizan una serie de nuevas propuestas metodológicas, que mejoran la eficiencia de las arquitecturas existentes, aplicadas a una serie de conjunto de datos reales que presenta un reto en la sociedad actual. La técnica conocida como deep learning ha adquirido gran popularidad en los últimos años debido a sus increíbles resultados en áreas como la visión artificial, procesamiento del lenguaje natural y predicción de series temporales, entre otras. Esta técnica se inspira en el funcionamiento de la célula básica del cerebro, la neurona. Las neuronas se organizan en capas formando una red neuronal, procesando la información de entrada y propagando su salida hacia otras capas de neuronas hasta obtener la salida final. Esta técnica ha sido adaptada en múltiples ocasiones a la predicción de series temporales desarrollando arquitecturas con unos resultados con resultados competitivos con el estado del arte actual. Sin embargo, aunque la eficacia ha sido un gran punto a favor, en ocasiones estas arquitecturas han degradado su eficiencia impidiendo su aplicación en escenarios reales. Existen diversas formas de mejorar la eficiencia, reduciendo algunos de los aspectos que toman gran cantidad de recursos como: memoria necesaria para almacenar la arquitectura, tiempo de inferencia o tiempo de entrenamiento, entre otros. Esta tesis se centra en mejorar el tiempo de entrenamiento, pues, resulta el cuello de botella a la hora de experimentar con nuevas arquitecturas, optimizar las arquitecturas existentes o reentrenar arquitecturas en ciertos escenarios reales. Ante el problema de eficiencia que presentan las arquitecturas dentro del ámbito del deep learning o las redes neuronales, se han desarrollado cuatro propuestas diferentes con el objetivo de obtener una eficacia igual o su

Time series is one of the most common data types in the industry nowadays. Forecasting the future of a time series behavior can be useful in planning ahead, saving time, resources, and helping avoid undesired scenarios. To make the forecasting, historical data is utilized due to the causal nature of the time series. Several deep learning algorithms have been presented in this area, where the input is processed through a series of non-linear functions to produce the output. We present a novel strategy to improve the performance of deep learning models in time series forecasting in terms of efficiency while reaching similar effectiveness. This approach separates the model into levels, starting with the easiest and continuing to the most difficult. The simpler levels deal with smoothed versions of the input, whereas the most sophisticated level deals with the raw data. This strategy seeks to mimic the human learning process, in which basic tasks are completed initially, followed by more precise and sophisticated ones. Our method achieved promising results, obtaining a 35% improvement in mean squared error and a 2.6 time decrease in training time compared with the best models found in a variety of time series.

This paper proposes a novel approach that combines an association rule algorithm with a deep learning model to enhance the interpretability of prediction outcomes. The study aims to gain insights into the patterns that were learned correctly or incorrectly by the model. To identify these scenarios, an association rule algorithm is applied to extract the patterns learned by the deep learning model. The rules are then analyzed and classified based on specific metrics to draw conclusions about the behavior of the model. We applied this approach to a well-known dataset in various scenarios, such as underfitting and overfitting. The results demonstrate that the combination of the two techniques is highly effective in identifying the patterns learned by the model and analyzing its performance in different scenarios, through error analysis. We suggest that this methodology can enhance the transparency and interpretability of black-box models, thus improving their reliability for real-world applications.

Traditional time series forecasting models often use all available variables, including potentially irrelevant or noisy features, which can lead to overfitting and poor performance. Feature selection can help address this issue by selecting the most informative variables in the temporal and feature dimensions. However, selecting the right features can be challenging for time series models. Embedded feature selection has been a popular approach, but many techniques do not include it in their design, including deep learning methods, which can lead to less efficient and effective feature selection. This paper presents a deep learning-based method for time series forecasting that incorporates feature selection to improve model efficacy and interpretability. The proposed method uses a multidimensional layer to remove irrelevant features along the temporal dimension. The resulting model is compared to several feature selection methods and experimental results demonstrate that the proposed approach can improve forecasting accuracy while reducing model complexity.

The current genomics era is bringing an unprecedented growth in the amount of gene expression data, only comparable to the exponential growth of sequences in databases during the last decades. This data allow the design of secondary analyses that take advantage of this information to create new knowledge. One of these feasible analyses is the evaluation of the expression level for a gene through a series of different conditions or cell types. Based on this idea, we have developed Automatic and Serial Analysis of CO-expression, which performs expression profiles for a given gene along hundreds of heterogeneous and normalized transcriptomics experiments and discover other genes that show either a similar or an inverse behavior. It might help to discover co-regulated genes, and common transcriptional regulators in any biological model. The present severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) pandemic is an opportunity to test this novel approach due to the wealth of data that are being generated, which could be used for validating results. Thus, we have identified 35 host factors in the literature putatively involved in the infectious cycle of SARS-CoV viruses and searched for genes tightly co-expressed with them. We have found 1899 co-expressed genes whose assigned functions are strongly related to viral cycles. Moreover, this set of genes heavily overlaps with those identified by former laboratory high-throughput screenings (with P-value near 0). Our results reveal a series of common regulators, involved in immune and inflammatory responses that might be key virus targets to induce the coordinated expression of SARS-CoV-2 host factors.

The current genomics era is bringing an unprecedented growth in the amount of gene expression data, only comparable to the exponential growth of sequences in databases during the last decades. This data now allows the design of secondary analyses that take advantage of this information to create new knowledge through specific computational approaches. One of these feasible analyses is the evaluation of the expression level for a gene through a series of different conditions or cell types. Based on this idea, we have developed ASACO, Automatic and Serial Analysis of CO-expression, which performs expression profiles for a given gene along hundreds of normalized and heterogeneous transcriptomics experiments and discover other genes that show either a similar or an inverse behavior. It might help to discover co-regulated genes, and even common transcriptional regulators in any biological model, including human diseases or microbial infections. The present SARS-CoV-2 pandemic is an opportunity to test this novel approach due to the wealth of data that is being generated, which could be used for validating results. In addition, new cell mechanisms identified could become new therapeutic targets. Thus, we have identified 35 host factors in the literature putatively involved in the infectious cycle of SARS-CoV and/or SARS-CoV-2 and searched for genes tightly co-expressed with them. We have found around 1900 co-expressed genes whose assigned functions are strongly related to viral cycles. Moreover, this set of genes heavily overlap with those identified by former laboratory high-throughput screenings (with p-value near 0). Some of these genes aim to cellular structures such as the stress granules, which could be essential for the virus replication and thereby could constitute potential targets in the current fight against the virus. Additionally, our results reveal a series of common transcription regulators, involved in immune and inflammatory responses, that might be key v

Purchase processes on Amazon Marketplace begin at the Buy Box, which represents the buy click process through which numerous sellers compete. This study aimed to estimate empirically the relevant seller characteristics that Amazon could consider featuring in the Buy Box. To that end, 22 product categories from Italy’s Amazon web page were studied over a ten-month period, and the sellers were analyzed through their products featured in the Buy Box. Two different experiments were proposed and the results were analyzed using four classification algorithms (a neural network, random forest, support vector machine, and C5.0 decision trees) and a rule-based classification. The first experiment aimed to characterize sellers unspecifically by predicting their change at the Buy Box. The second one aimed to predict which seller would be featured in it. Both experiments revealed that the customer experience and the dynamics of the sellers’ prices were important features of the Buy Box. Additionally, we proposed a set of default features that Amazon could consider when no information about sellers was available. We also proposed the possible existence of a relationship or composition among important features that could be used for sellers to be featured in the Buy Box.

The agricultural sector has been, and still is, the most important economic sector in many countries. Due to advances in technology, the amount and variety of available data have been increasing over the years. However, compared to other economic sectors, there is not always enough quality data for one particular domain (crops, plantations, plots) to obtain acceptable forecasting results with machine learning algorithms. In this context, transfer learning can help extract knowledge from different but related domains with enough data to transfer it to a target domain with scarce data. This process can overcome forecasting accuracy compared to training models uniquely with data from the target domain. In this work, a novel instance weighting-based transfer learning algorithm is proposed and applied to the phenology forecasting problem. A new metric named DIAFAN is proposed to weight samples from different source domains according to their relationship with the target domain, promoting the diversity of the information and avoiding inconsistent samples. Additionally, a set of validation schemes is specifically designed to ensure fair comparisons in terms of data volume with other benchmark transfer learning algorithms. The proposed algorithm, DIAFAN-TL, is tested with a proposed dataset of 16 plots of olive groves from different places, including information fusion from satellite images, meteorological stations and human field sampling of crop phenology. DIAFAN-TL achieves a remarkable improvement with respect to 15 other well-known transfer learning algorithms and three nontransfer learning scenarios. Finally, several performance analyses according to the different phenological states, prediction horizons and source domains are also performed.

Neural networks have proven to be a good alternative in application fields such as healthcare, time-series forecasting and artificial vision, among others, for tasks like regression or classification. Their potential has been particularly remarkable in unstructured data, but recently developed architectures or their ensemble with other classical methods have produced competitive results in structured data. Feature selection has several beneficial properties: improve efficacy, performance, problem understanding and data recollection time. However, as new data sources become available and new features are generated using feature engineering techniques, more computational resources are required for feature selection methods. Feature selection takes an exorbitant amount of time in datasets with numerous features, making it impossible to use or achieving suboptimal selections that do not reflect the underlying behavior of the problem. We propose a nonparametric neural network layer which provides all the benefits of feature selection while requiring few changes to the architecture. Our method adds a novel layer at the beginning of the neural network, which removes the influence of features during training, adding inherent interpretability to the model without extra parameterization. In contrast to other feature selection methods, we propose an efficient and model-aware method to select the features with no need to train the model several times. We compared our method with a variety of popular feature selection strategies and datasets, showing remarkable results.

This study proposes a novel bioinspired metaheuristic simulating how the coronavirus spreads and infects healthy people. From a primary infected individual (patient zero), the coronavirus rapidly infects new victims, creating large populations of infected people who will either die or spread infection. Relevant terms such as reinfection probability, super-spreading rate, social distancing measures, or traveling rate are introduced into the model to simulate the coronavirus activity as accurately as possible. The infected population initially grows exponentially over time, but taking into consideration social isolation measures, the mortality rate, and number of recoveries, the infected population gradually decreases. The coronavirus optimization algorithm has two major advantages when compared with other similar strategies. First, the input parameters are already set according to the disease statistics, preventing researchers from initializing them with arbitrary values. Second, the approach has the ability to end after several iterations, without setting this value either. Furthermore, a parallel multivirus version is proposed, where several coronavirus strains evolve over time and explore wider search space areas in less iterations. Finally, the metaheuristic has been combined with deep learning models, to find optimal hyperparameters during the training phase. As application case, the problem of electricity load time series forecasting has been addressed, showing quite remarkable performance.

A new transfer learning strategy is proposed for image classification in this work, based on an 8-layer convolutional neural network. The transfer learning process consists in a training phase of the neural network on a source dataset of images. Then, the last two layers are retrained using a different small target dataset of images. A preliminary study was conducted to train and test the transfer learning proposal on Malaria cell images for a binary classification problem. The methodology proposed has provided a 6.76% of improvement with respect to other three different strategies of training non-transfer learning models. The results achieved are quite promising and encourage to conduct further research in this field.

Surface ozone (O3) is considered an hazard to human health, affecting vegetation crops and ecosystems. Accurate time and location O3 forecasting can help to protect citizens to unhealthy exposures when high levels are expected. Usually, forecasting models use numerous O3 precursors as predictors, limiting the reproducibility of these models to the availability of such information from data providers. This study introduces a 24 h-ahead hourly O3 concentrations forecasting methodology based on bagging and ensemble learning, using just two predictors with lagged O3 concentrations. This methodology was applied on ten-year time series (2006–2015) from three major urban areas of Andalusia (Spain). Its forecasting performance was contrasted with an algorithm especially designed to forecast time series exhibiting temporal patterns. The proposed methodology outperforms the contrast algorithm and yields comparable results to others existing in literature. Its use is encouraged due to its forecasting performance and wide applicability, but also as benchmark methodology.