Alejandro Carrasco

Non-Intrusive Load Monitoring (NILM) aims to identify electrical appliances in a household from voltage and current signals, enabling energy efficiency and smart grid applications. This study investigates NILM classification using the PLAID sub-metered dataset, evaluating multiple feature extraction techniques and classification algorithms. The results demonstrate that generic statistical features outperform domain-specific electrical features, including those based on IEEE standards. Feature analysis reveals that transient-state signals generally lead to higher classification accuracy than steady-state signals. However, the best performance is achieved by combining statistical features from both states, specifically by characterizing the steady-state current signal, complemented by its phase shift, and the active-state current signal filtered through nine pass-band filters. Dimensionality reduction further improves classification by enhancing feature relevance. Among the classifiers tested, Random Forest achieves the highest performance, reaching 93% accuracy under optimal conditions. To address dataset imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) was applied. While this technique does not improve overall performance, it enhances classification fairness by improving recognition of underrepresented categories at the cost of a slight decrease in the most populated ones. Learning curve analysis suggests that performance has not yet plateaued, indicating that a larger dataset would further enhance results. These findings highlight key challenges and opportunities in NILM, suggesting that future work should focus on expanding dataset diversity and refining feature extraction methods to improve classification robustness.

Bottling machinery is a critical component in agri-food industries, where maintaining operational efficiency is key to ensuring productivity and minimizing economic losses. Early detection of faulty conditions in this equipment can significantly improve maintenance procedures and overall system performance. This research focuses on health monitoring of gripping pliers in bottling plants, a crucial task that has traditionally relied on analyzing raw vibration signals or using narrowly defined, application-specific features. However, these methods often face challenges related to limited robustness, high computational costs, and sensitivity to noise. To address these limitations, we propose a novel approach based on generic features extracted through basic signal processing techniques applied to vibration signals. These features are then classified using a random forest algorithm, enabling an effective analysis of health states. The proposed method is evaluated against traditional approaches and demonstrates clear advantages, including higher accuracy in detecting and classifying faulty conditions, greater robustness against random perturbations, and a reduced computational cost. Additionally, the method requires fewer training instances to achieve reliable performance. This study highlights the potential of artificial intelligence and signal processing techniques in predictive maintenance, offering a scalable and efficient solution for fault detection in manufacturing processes, particularly within the agri-food sector.

In this article a methodology, based on the Kano model, to prioritize the features of a product or service is proposed. Instead of using detailed, lengthy, burdensome, time-demanding, and biased-prone questionnaires, enquiring the user satisfaction with each product feature, a simplified survey asking for the overall satisfaction with the product is used. The proposed method starts by training a machine learning (ML) model using a dataset of different instances of the product and the corresponding perceived quality. This model is then employed to derive the relationship between each feature and the satisfaction associated with them. The shape of this relationship is interpreted according to a Kano model placing each attribute in a bidimensional Kano map which is later partitioned using ML clustering techniques. This methodology has been applied to an open dataset containing the physicochemical characteristics of hundreds of wines and the corresponding scores obtained in a blind tasting evaluation. The research has shown that ML models get very remarkable results predicting the perceived quality of a wine and is able to build a Kano map of the wine features. The ML clustering techniques employed partitioning this Kano map has clearly overperformed conventional rectangular or polar segmentation. It has also been shown that using four categories of features, as it is proposed in the Kano model, is the most reasonable partition from an ML clustering perspective.

Predictive maintenance (PdM) is a set of actions and techniques to early detect failures and defects on machines before they occur, and the usage of machine learning and deep learning techniques in predictive maintenance has increased during the last years. Even with this increase of the literature, there is still a gap concerning the application of such techniques for PdM in the industry, as there are no clear guidelines about which information to use for a PdM system, how to process the information, and what machine learning techniques should be used in order to obtain acceptable results. This scoping review is performed in order to observe the current status on the use of Machine Learning and Deep Learning in predictive maintenance in academia and provide answer to the questions related to these guidelines. For this purpose, a literature review of the last five years is carried out, using those articles that cover information about sources of information used for PdM, the treatment given to such data and the machine learning (ML) methods or techniques used. The Web of Science: Core Collection database is used as a source of information, specifically the Science Citation Index Expanded (SCIE). The review shows that there are different information sources used for machine learning and deep learning in PdM, depending on the origin of the data and the availability of it, and as well whether the data sets are private or public. Also, we can observe that data used for both training and making predictions does not only use traditional pre-processing techniques, but that about one fifth of the articles even propose new techniques in this field. Additionally, we compare a wide range of techniques and algorithms which are used in Deep Learning -being ANN the most used- and in Machine Learning, being SVM the most used algorithm, closely followed by Random Forest. Based on the results, we provide indications about how to apply ML for PdM in industry. Keywords: machine learning, predictive maintenance, artificial intelligence, deep learning, data processing, data collection

In electrical distribution networks an adequate management is key for supporting the deployment of renewable generation sources and microgrids while extracting their maximum potential. Among the existing optimization approaches, stochastic and probabilistic methods are experiencing a growth in their use. However, one of the problems when applying these approaches is the complexity of creating and evaluating the quality of the required stochastic forecasts compared to deterministic forecasts. To mitigate this difficulty, this paper proposes a probabilistic forecasting framework that integrates model creation, their evaluation, and the selection of the best model for predicting. Additionally, two novel methods are proposed for creating scenario sets, and a new metric is defined for evaluating and selecting which model to use. The proposed framework is applied in a case study over a dataset of ten secondary distribution substations from a real distribution network located in Manzanilla (Spain), showing the effect of the selection criteria over the forecasting quality. Note to Practitioners —This article was motivated by the challenge of probabilistic forecasting inclusion in automatic management systems applied to power distribution networks and microgrids. Modern stochastic management optimization methods are fed with probabilistic forecasts, which offer richer information than classic deterministic forecasting. Therefore, the management systems should be able to automatically train a certain number of forecasting models (e.g., machine learning models), evaluate and compare them, and apply the best ones for obtaining the forecasts to feed the management optimization system. Considering the variety of models, techniques, probabilistic forecast types, and evaluation metrics, it can be unclear how to perform this process. For these reasons, this article proposes a probabilistic forecasting framework that integrates methods for the construction of diverse types of predictions (quantiles, intervals, and scenario sets), their evaluation, and the selection of the best model for performing each required prediction for feeding optimization systems. This framework could help to facilitate the implantation of modern stochastic optimization management systems for distribution networks and microgrids, as it simplifies the forecasting process.

The increasing competition in the electric sector is challenging retail companies as they must assign its commercial efforts to attract the most profitable customers. Those are whose energy demand best fit certain target profiles, which usually depend on generation or cost policies. But, even when the demand profile is available, it is in an anonymous way, preventing its association to a particular client. In this paper, we explore a large dataset containing several millions of monthly demand profiles in Spain and use the available information about the associated economic sector and location for an indirect identification of the customers. The distance of the demand profile from the target is used to define a key performance indicator (KPI) which is used as the main driver of the proposed marketing strategy. The combined use of activity and location has been revealed as a powerful tool for indirect identification of customers, as 100,000 customers are uniquely identified, while about 300,000 clients are identifiable in small sets containing 10 or less consumers. To assess the proposed marketing strategy, it has been compared to the random attraction of new clients, showing a reduction of distance from the target of 40% for 10,000 new customers.

Recent developments in machine learning applications are deeply concerned with the poor interpretability of most of these techniques. To gain some insights in the process of designing data-based models it is common to graphically represent the algorithm’s results, either in their final or intermediate stage. Specially challenging is the task of plotting multiclass classification results as they involve categorical variables (classes) rather than numeric results. Using the well-known MNIST dataset and a simple neural network as an example, this paper reviews the existing techniques to visualize classification results, from those centered on a particular instance or set of instances, to those representing an overall performance metric. As classification results are commonly summarized in the form of a confusion matrix, special attention is paid to its graphical representation. From this analysis, a new visualization tool is derived, which is presented in two forms: confusion star and confusion gear. The confusion star is centered on the classification errors, while the confusion gear focuses on the classification hits. The proposed visualization tools are also evaluated when facing: (i) balanced and imbalanced classifiers issues; (ii) the problem of representing errors with different orders of magnitude. By using shapes instead of colors to represent the value of each matrix cell, the new tools significantly improve the readability of the confusion matrices. Furthermore, we show how the area enclosed by the confusion stars and gears are directly related to standard classification metrics. The new graphic tools can be also usefully employed to visualize the performances of a sequence of classifiers.

Background: Cancer patients are increasingly using mobile health (mHealth) apps to take control of their health. Many studies have explored their efficiency, content, usability, and adherence; however, these apps have created a new set of privacy challenges, as they store personal and sensitive data. Objective: The purpose of this study was to refine and evaluate a scale based on the General Data Protection Regulation and assess the fairness of privacy policies of mHealth apps. Methods: Based on the experience gained from our previous work, we redefined some of the items and scores of our privacy scale. Using the new version of our scale, we conducted a case study in which we analyzed the privacy policies of cancer Android apps. A systematic search of cancer mobile apps was performed in the Spanish version of the Google Play website. Results: The redefinition of certain items reduced discrepancies between reviewers. Thus, use of the scale was made easier, not only for the reviewers but also for any other potential users of our scale. Assessment of the privacy policies revealed that 29% (9/31) of the apps included in the study did not have a privacy policy, 32% (10/31) had a score over 50 out of a maximum of 100 points, and 39% (12/31) scored fewer than 50 points. Conclusions: In this paper, we present a scale for the assessment of mHealth apps that is an improved version of our previous scale with adjusted scores. The results showed a lack of fairness in the mHealth app privacy policies that we examined, and the scale provides developers with a tool to evaluate their privacy policies.

Background: Privacy has always been a concern, especially in the health domain. The proliferation of mobile health (mHealth) apps has led to a large amount of sensitive data being generated. Some authors have performed privacy assessments of mHealth apps. They have evaluated diverse privacy components; however, different authors have used different criteria for their assessments. Objective: This scoping review aims to understand how privacy is assessed for mHealth apps, focusing on the components, scales, criteria, and scoring methods used. A simple taxonomy to categorize the privacy assessments of mHealth apps based on component evaluation is also proposed. Methods: We followed the methodology defined by Arksey and O’Malley to conduct a scoping review. Included studies were categorized based on the privacy component, which was assessed using the proposed taxonomy. Results: The database searches retrieved a total of 710 citations—24 of them met the defined selection criteria, and data were extracted from them. Even though the inclusion criteria considered articles published since 2009, all the studies that were ultimately included were published from 2014 onward. Although 12 papers out of 24 (50%) analyzed only privacy, 8 (33%) analyzed both privacy and security. Moreover, 4 papers (17%) analyzed full apps, with privacy being just part of the assessment. The evaluation criteria used by authors were heterogeneous and were based on their experience, the literature, and/or existing legal frameworks. Regarding the set of items used for the assessments, each article defined a different one. Items included app permissions, analysis of the destination, analysis of the content of communications, study of the privacy policy, use of remote storage, and existence of a password to access the app, among many others. Most of the included studies provided a scoring method that enables the comparison of privacy among apps. Conclusions: The privacy assessment of mHealth apps is a complex task, as the criteria used by different authors for their evaluations are very heterogeneous. Although some studies about privacy assessment have been conducted, a very large set of items to evaluate privacy has been used up until now. In-app information and privacy policies are primarily utilized by the scientific community to extract privacy information from mHealth apps. The creation of a scale based on more objective criteria is a desirable step forward for privacy assessment in the future.

BACKGROUND Privacy has always been a concern, especially in the health domain. The proliferation of mobile health (mHealth) apps has led to a large amount of sensitive data being generated. Some authors have performed privacy assessments of mHealth apps. They have evaluated diverse privacy components; however, different authors have used different criteria for their assessments. OBJECTIVE This scoping review aims to understand how privacy is assessed for mHealth apps, focusing on the components, scales, criteria, and scoring methods used. A simple taxonomy to categorize the privacy assessments of mHealth apps based on component evaluation is also proposed. METHODS We followed the methodology defined by Arksey and O’Malley to conduct a scoping review. Included studies were categorized based on the privacy component, which was assessed using the proposed taxonomy. RESULTS The database searches retrieved a total of 710 citations—24 of them met the defined selection criteria, and data were extracted from them. Even though the inclusion criteria considered articles published since 2009, all the studies that were ultimately included were published from 2014 onward. Although 12 papers out of 24 (50%) analyzed only privacy, 8 (33%) analyzed both privacy and security. Moreover, 4 papers (17%) analyzed full apps, with privacy being just part of the assessment. The evaluation criteria used by authors were heterogeneous and were based on their experience, the literature, and/or existing legal frameworks. Regarding the set of items used for the assessments, each article defined a different one. Items included app permissions, analysis of the destination, analysis of the content of communications, study of the privacy policy, use of remote storage, and existence of a password to access the app, among many others. Most of the included studies provided a scoring method that enables the comparison of privacy among apps. CONCLUSIONS The privacy assessment of mHealth apps is a complex task, as the criteria used by different authors for their evaluations are very heterogeneous. Although some studies about privacy assessment have been conducted, a very large set of items to evaluate privacy has been used up until now. In-app information and privacy policies are primarily utilized by the scientific community to extract privacy information from mHealth apps. The creation of a scale based on more objective criteria is a desirable step forward for privacy assessment in the future.