Abstract
To maintain the performance of the latest generation of onshore and offshore wind turbine systems, a new methodology must be proposed to enhance the maintenance policy. In this context, this paper introduces an approach to designing a decision support tool that combines predictive capabilities with anomaly explanations for effective IoT predictive maintenance tasks. Essentially, the paper proposes an approach that integrates a predictive maintenance model with an explicative decision-making system. The key challenge is to detect anomalies and provide plausible explanations, enabling human operators to determine the necessary actions swiftly. To achieve this, the proposed approach identifies a minimal set of relevant features required to generate rules that explain the root causes of issues in the physical system. It estimates that certain features, such as the active power generator, blade pitch angle, and the average water temperature of the voltage circuit protection in the generator's sub-components, are particularly critical to monitor. Additionally, the approach simplifies the computation of an efficient predictive maintenance model. Compared to other deep learning models, the identified model provides up to 80% accuracy in anomaly detection and up to 96% for predicting the remaining useful life of the system under study. These performance metrics and indicators values are essential for enhancing the decision-making process. Moreover, the proposed decision support tool elucidates the onset of degradation and its dynamic evolution based on expert knowledge and data gathered through Internet of Things (IoT) technology and inspection reports. Thus, the developed approach should aid maintenance managers in making accurate decisions regarding inspection, replacement, and repair tasks. The methodology is demonstrated using a wind farm dataset provided by Energias De Portugal.

Abstract
Data-driven decision models based on Artificial Intelligence (AI) have been widely used in the public and private sectors. These models present challenges and are intended to be fair, effective and transparent in public interest areas. Bias, fairness and government transparency are aspects that significantly impact the functioning of a democratic society. They shape the government's and its citizens' relationship, influencing trust, accountability, and the equitable treatment of individuals and groups. Data-driven decision models can be biased at several process stages, contributing to injustices. Our research purpose is to understand fairness in the use of causal discovery for public procurement. By analysing Portuguese public contracts data, we aim i) to predict the place of execution of public contracts using the PC algorithm with sp-mi, smc-chi(2) and mc-chi(2) conditional independence tests; ii) to analyse and compare the fairness in those scenarios using Predictive Parity Rate, Proportional Parity, Demographic Parity and Accuracy Parity metrics. By addressing fairness concerns, we pursue to enhance responsible data-driven decision models. We conclude that, in our case, fairness metrics make an assessment more local than global due to causality pathways. We also observe that the Proportional Parity metric is the one with the lowest variance among all metrics and one with the highest precision, and this reinforces the observation that the Agency category is the one that is furthest apart in terms of the proportion of the groups.

Abstract
In many data stream applications, there is a normal concept, and the objective is to identify normal and abnormal concepts by training only with normal concept instances. This scenario is known in the literature as one-class learning (OCL) for data streams. In this OCL scenario for data streams, we highlight two main gaps: (i) lack of methods based on graph neural networks (GNNs) and (ii) lack of interpretable methods. We introduce OPENCAST (One-class graPh autoENCoder for dAta STream), a new method for data streams based on OCL and GNNs. Our method learns representations while encapsulating the instances of interest through a hypersphere. OPENCAST learns low-dimensional representations to generate interpretability in the representation learning process. OPENCAST achieved state-of-the-art results for data streams in the OCL scenario, outperforming seven other methods. Furthermore, OPENCAST learns low-dimensional representations, generating interpretability in the representation learning process and results.

Abstract
Latent Dirichlet Allocation (LDA) is a fundamental method for clustering short text streams. However, when applied to large datasets, it often faces significant challenges, and its performance is typically evaluated in domain-specific datasets such as news and tweets. This study aims to fill this gap by evaluating the effectiveness of short text clustering methods in a large and diverse e-commerce dataset. We specifically investigate how well these clustering algorithms adapt to the complex dynamics and larger scale of e-commerce text streams, which differ from their usual application domains. Our analysis focuses on the impact of high homogeneity scores on the reported Normalized Mutual Information (NMI) values. We particularly examine whether these scores are inflated due to the prevalence of single-element clusters. To address potential biases in clustering evaluation, we propose using the Akaike Information Criterion (AIC) as an alternative metric to reduce the formation of single-element clusters and provide a more balanced measure of clustering performance. We present new insights for applying short text clustering methodologies in real-world situations, especially in sectors like e-commerce, where text data volumes and dynamics present unique challenges.

Abstract

Abstract