José Saias
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
José Saias has a total of 38 co-authors. Here are the top co-authors by number of shared publications:
- Paulo Quaresma
University of Évora
13 shared publications - Vitor Nogueira
University of Évora
7 shared publications - Daniel Santos
University of Évora
6 shared publications - Anabela Afonso
Centro de Investigação em Matemática e Aplicações (CIMA), Universidade de Évora, Évora, Portugal, uevora.pt
5 shared publications - Gonçalo Jacinto
University of Évora
5 shared publications - Leonor Rego
Department of Mathematics, ECT, University of Évora, 7000-671 Évora, Portugal
5 shared publications
Latest Publications
- Open Access
- 24 June 2025
- Published by MDPI AG in Electronics
- Vol. 14 (13) , 2551
Advances in NLP Techniques for Detection of Message-Based Threats in Digital Platforms: A Systematic Review
Users of all ages face risks on social media and messaging platforms. When encountering suspicious messages, legitimate concerns arise about a sender’s malicious intent. This study examines recent advances in Natural Language Processing for detecting message-based threats in digital communication. We conducted a systematic review following PRISMA guidelines, to address four research questions. After applying a rigorous search and screening pipeline, 30 publications were selected for analysis. Our work assessed the NLP techniques and evaluation methods employed in recent threat detection research, revealing that large language models appear in only 20% of the reviewed works. We further categorized detection input scopes and discussed ethical and privacy implications. The results show that AI ethical aspects are not systematically addressed in the reviewed scientific literature.
- 24 June 2025
- Published by MDPI AG in Electronics
- Vol. 14 (13) , 2551
Open AccessAdvances in NLP Techniques for Detection of Message-Based Threats in Digital Platforms: A Systematic Review
Users of all ages face risks on social media and messaging platforms. When encountering suspicious messages, legitimate concerns arise about a sender’s malicious intent. This study examines recent advances in Natural Language Processing for detecting message-based threats in digital communication. We conducted a systematic review following PRISMA guidelines, to address four research questions. After applying a rigorous search and screening pipeline, 30 publications were selected for analysis. Our work assessed the NLP techniques and evaluation methods employed in recent threat detection research, revealing that large language models appear in only 20% of the reviewed works. We further categorized detection input scopes and discussed ethical and privacy implications. The results show that AI ethical aspects are not systematically addressed in the reviewed scientific literature.
- Open Access
- 17 September 2024
- Published by MDPI AG in Electronics
- Vol. 13 (18) , 3687
Sensor-Based Real-Time Monitoring Approach for Multi-Participant Workout Intensity Management
One of the significant advantages of technological evolution is the greater ease of collecting and analyzing data. Miniaturization, wireless communication protocols and IoT allow the use of sensors to collect data, with all the potential to support decision making in real time. In this paper, we describe the design and implementation of a digital solution to guide the intensity of training or physical activity, based on heart rate wearable sensors applied to participants in group sessions. Our system, featuring a unified engine that simplifies sensor management and minimizes user disruption, has been proven effective for real-time monitoring. It includes custom alerts during variable-intensity workouts, and ensures data preservation for subsequent analysis by physiologists or clinicians. This solution has been used in sessions of up to six participants and sensors up to 12 m away from the gateway device. We describe some challenges and constraints we face in collecting data from multiple and possibly different sensors simultaneously via Bluetooth Low Energy, and the approaches we follow to overcome them. We conduct an in-depth questionnaire to identify potential obstacles and drivers for system acceptance. We also discuss some possibilities for extension and improvement of our system.
- 17 September 2024
- Published by MDPI AG in Electronics
- Vol. 13 (18) , 3687
Open AccessSensor-Based Real-Time Monitoring Approach for Multi-Participant Workout Intensity Management
One of the significant advantages of technological evolution is the greater ease of collecting and analyzing data. Miniaturization, wireless communication protocols and IoT allow the use of sensors to collect data, with all the potential to support decision making in real time. In this paper, we describe the design and implementation of a digital solution to guide the intensity of training or physical activity, based on heart rate wearable sensors applied to participants in group sessions. Our system, featuring a unified engine that simplifies sensor management and minimizes user disruption, has been proven effective for real-time monitoring. It includes custom alerts during variable-intensity workouts, and ensures data preservation for subsequent analysis by physiologists or clinicians. This solution has been used in sessions of up to six participants and sensors up to 12 m away from the gateway device. We describe some challenges and constraints we face in collecting data from multiple and possibly different sensors simultaneously via Bluetooth Low Energy, and the approaches we follow to overcome them. We conduct an in-depth questionnaire to identify potential obstacles and drivers for system acceptance. We also discuss some possibilities for extension and improvement of our system.
- Open Access
- 29 October 2023
- Published by MDPI AG in Electronics
- Vol. 12 (21) , 4447
Detecting Persuasion Attempts on Social Networks: Unearthing the Potential of Loss Functions and Text Pre-Processing in Imbalanced Data Settings
The rise of social networks and the increasing amount of time people spend on them have created a perfect place for the dissemination of false narratives, propaganda, and manipulated content. In order to prevent the spread of disinformation, content moderation is needed. However, manual moderation is unfeasible due to the large amount of daily posts. This paper studies the impact of using different loss functions on a multi-label classification problem with an imbalanced dataset, consisting of 20 persuasion techniques and only 950 samples, provided by SemEval’s 2021 Task 6. We used machine learning models, such as Naive Bayes and Decision Trees, and a custom deep learning architecture, based on DistilBERT and Convolutional Layers. Overall, the machine learning models achieved far worse results than the deep learning model, using Binary Cross Entropy, which we considered our baseline deep learning model. To address the class imbalance problem, we trained our model using different loss functions, such as Focal Loss and Asymmetric Loss. The latter providing the best results, particularly for the least represented classes.
- 29 October 2023
- Published by MDPI AG in Electronics
- Vol. 12 (21) , 4447
Open AccessDetecting Persuasion Attempts on Social Networks: Unearthing the Potential of Loss Functions and Text Pre-Processing in Imbalanced Data Settings
The rise of social networks and the increasing amount of time people spend on them have created a perfect place for the dissemination of false narratives, propaganda, and manipulated content. In order to prevent the spread of disinformation, content moderation is needed. However, manual moderation is unfeasible due to the large amount of daily posts. This paper studies the impact of using different loss functions on a multi-label classification problem with an imbalanced dataset, consisting of 20 persuasion techniques and only 950 samples, provided by SemEval’s 2021 Task 6. We used machine learning models, such as Naive Bayes and Decision Trees, and a custom deep learning architecture, based on DistilBERT and Convolutional Layers. Overall, the machine learning models achieved far worse results than the deep learning model, using Binary Cross Entropy, which we considered our baseline deep learning model. To address the class imbalance problem, we trained our model using different loss functions, such as Focal Loss and Asymmetric Loss. The latter providing the best results, particularly for the least represented classes.
- Open Access
- 13 April 2023
- Published by MDPI AG in Information
- Vol. 14 (4) , 238
Prediction of Road Traffic Accidents on a Road in Portugal: A Multidisciplinary Approach Using Artificial Intelligence, Statistics, and Geographic Information Systems
Road Traffic Accidents (RTA) cause human losses and irreparable physical and psychological damage to many of the victims. They also involve a very relevant economic dimension. It is urgent to improve the management of human and material resources for more effective prevention. This work makes an important contribution by presenting a methodology that allowed for achieving a predictive model for the occurrence of RTA on a road with a high RTA rate. The prediction is obtained for each road segment for a given time and day and combines results from statistical methods, spatial analysis, and artificial intelligence models. The performance of three Machine Learning (ML) models (Random Forest, C5.0 and Logistic Regression) is compared using different approaches for imbalanced data (random sampling, directional sampling, and Random Over-Sampling Examples (ROSE)) and using different segment lengths (500 m and 2000 m). This study used RTA data from 2016–2019 (training) and from May 2021–June 2022 (test). The most effective model was an ML logistic regression with the ROSE approach, using segments length 500 m (sensitivity = 87%, specificity = 60%, AUC = 0.82). The model was implemented in a digital application, and a Portuguese security force is already using it.
- 13 April 2023
- Published by MDPI AG in Information
- Vol. 14 (4) , 238
Open AccessPrediction of Road Traffic Accidents on a Road in Portugal: A Multidisciplinary Approach Using Artificial Intelligence, Statistics, and Geographic Information Systems
Road Traffic Accidents (RTA) cause human losses and irreparable physical and psychological damage to many of the victims. They also involve a very relevant economic dimension. It is urgent to improve the management of human and material resources for more effective prevention. This work makes an important contribution by presenting a methodology that allowed for achieving a predictive model for the occurrence of RTA on a road with a high RTA rate. The prediction is obtained for each road segment for a given time and day and combines results from statistical methods, spatial analysis, and artificial intelligence models. The performance of three Machine Learning (ML) models (Random Forest, C5.0 and Logistic Regression) is compared using different approaches for imbalanced data (random sampling, directional sampling, and Random Over-Sampling Examples (ROSE)) and using different segment lengths (500 m and 2000 m). This study used RTA data from 2016–2019 (training) and from May 2021–June 2022 (test). The most effective model was an ML logistic regression with the ROSE approach, using segments length 500 m (sensitivity = 87%, specificity = 60%, AUC = 0.82). The model was implemented in a digital application, and a Portuguese security force is already using it.
- Open Access
- 24 February 2023
- Published by MDPI AG in ISPRS International Journal of Geo-Information
- Vol. 12 (3) , 93
Learning from Accidents: Spatial Intelligence Applied to Road Accidents with Insights from a Case Study in Setúbal District, Portugal
Road traffic accidents are a major concern for modern society with a high toll on human life and involve hard to account economic consequences. New knowledge can be obtained from combining GIS tools with machine learning and artificial intelligence, developing what is, in this work, identified as spatial intelligence. This approach is tested in a case study of Setúbal district, Portugal, for the period of 2016 to 2019. Departing from a heatmap analysis, and applying kernel density estimation, new spatial approaches were used, namely DBSCAN and Getis-Ord. The results obtained allowed the identification of novel meaningful locations of road traffic accidents. Consequently, the knowledge built from the underlying patterns is considered the key to developing new strategies to solve this modern social curse. The methodology proposed in this study demonstrates that the combination of expertise built from the different spatial analyses can provide a better understanding of the determinants of road traffic accidents. This approach is expected to be valuable for data analysts and decision-makers, contributing to diminishing human losses related to road traffic accidents.
- 24 February 2023
- Published by MDPI AG in ISPRS International Journal of Geo-Information
- Vol. 12 (3) , 93
Open AccessLearning from Accidents: Spatial Intelligence Applied to Road Accidents with Insights from a Case Study in Setúbal District, Portugal
Road traffic accidents are a major concern for modern society with a high toll on human life and involve hard to account economic consequences. New knowledge can be obtained from combining GIS tools with machine learning and artificial intelligence, developing what is, in this work, identified as spatial intelligence. This approach is tested in a case study of Setúbal district, Portugal, for the period of 2016 to 2019. Departing from a heatmap analysis, and applying kernel density estimation, new spatial approaches were used, namely DBSCAN and Getis-Ord. The results obtained allowed the identification of novel meaningful locations of road traffic accidents. Consequently, the knowledge built from the underlying patterns is considered the key to developing new strategies to solve this modern social curse. The methodology proposed in this study demonstrates that the combination of expertise built from the different spatial analyses can provide a better understanding of the determinants of road traffic accidents. This approach is expected to be valuable for data analysts and decision-makers, contributing to diminishing human losses related to road traffic accidents.
- Open Access
- 27 January 2023
- Published by MDPI AG in Sustainability
- Vol. 15 (3) , 2352
Factors That Influence the Type of Road Traffic Accidents: A Case Study in a District of Portugal
Road traffic accidents (RTAs) are a problem with repercussions in several dimensions: social, economic, health, justice, and security. Data science plays an important role in its explanation and prediction. One of the main objectives of RTA data analysis is to identify the main factors associated with a RTA. The present study aims to contribute to the identification of the determinants for the type of RTA: collision, crash, or pedestrian running-over. These factors are essential for identifying specific countermeasures because there is a relevant relationship between the type of RTA and its severity. Daily RTA data from 2016 to 2019 in a district of Portugal were analyzed. A statistical multinomial logit model was fitted. The identified determinants for the type of RTA were geographical (municipality, location, and parking areas), meteorological (air temperature and weather), time of the day (hour, day of the week, and month), driver’s characteristics (gender and age), vehicle’s features (type and age) and road characteristics (road layout and type). The multinomial model results were compared with several machine learning algorithms, since the original data of the type of RTA are severely imbalanced. All models showed poor performance. However, when combining these models with ROSE for class balancing, their performance improved considerably, with the random forest algorithm showing the best performance.
- 27 January 2023
- Published by MDPI AG in Sustainability
- Vol. 15 (3) , 2352
Open AccessFactors That Influence the Type of Road Traffic Accidents: A Case Study in a District of Portugal
Road traffic accidents (RTAs) are a problem with repercussions in several dimensions: social, economic, health, justice, and security. Data science plays an important role in its explanation and prediction. One of the main objectives of RTA data analysis is to identify the main factors associated with a RTA. The present study aims to contribute to the identification of the determinants for the type of RTA: collision, crash, or pedestrian running-over. These factors are essential for identifying specific countermeasures because there is a relevant relationship between the type of RTA and its severity. Daily RTA data from 2016 to 2019 in a district of Portugal were analyzed. A statistical multinomial logit model was fitted. The identified determinants for the type of RTA were geographical (municipality, location, and parking areas), meteorological (air temperature and weather), time of the day (hour, day of the week, and month), driver’s characteristics (gender and age), vehicle’s features (type and age) and road characteristics (road layout and type). The multinomial model results were compared with several machine learning algorithms, since the original data of the type of RTA are severely imbalanced. All models showed poor performance. However, when combining these models with ROSE for class balancing, their performance improved considerably, with the random forest algorithm showing the best performance.
- 29 November 2022
- Published by Springer Nature
- p. 203-214
Some Determinants for Road Accidents Severity in the District of Setúbal
- 29 November 2022
- Published by Springer Nature
- p. 203-214
Some Determinants for Road Accidents Severity in the District of Setúbal
- Open Access
Comparison of Statistical and Machine-Learning Models on Road Traffic Accident Severity Classification
Portugal has the sixth highest road fatality rate among European Union members. This is a problem of different dimensions with serious consequences in people’s lives. This study analyses daily data from police and government authorities on road traffic accidents that occurred between 2016 and 2019 in a district of Portugal. This paper looks for the determinants that contribute to the existence of victims in road traffic accidents, as well as the determinants for fatalities and/or serious injuries in accidents with victims. We use logistic regression models, and the results are compared to the machine-learning model results. For the severity model, where the response variable indicates whether only property damage or casualties resulted in the traffic accident, we used a large sample with a small imbalance. For the serious injuries model, where the response variable indicates whether or not there were victims with serious injuries and/or fatalities in the traffic accident with victims, we used a small sample with very imbalanced data. Empirical analysis supports the conclusion that, with a small sample of imbalanced data, machine-learning models generally do not perform better than statistical models; however, they perform similarly when the sample is large and has a small imbalance.
Open AccessComparison of Statistical and Machine-Learning Models on Road Traffic Accident Severity Classification
Portugal has the sixth highest road fatality rate among European Union members. This is a problem of different dimensions with serious consequences in people’s lives. This study analyses daily data from police and government authorities on road traffic accidents that occurred between 2016 and 2019 in a district of Portugal. This paper looks for the determinants that contribute to the existence of victims in road traffic accidents, as well as the determinants for fatalities and/or serious injuries in accidents with victims. We use logistic regression models, and the results are compared to the machine-learning model results. For the severity model, where the response variable indicates whether only property damage or casualties resulted in the traffic accident, we used a large sample with a small imbalance. For the serious injuries model, where the response variable indicates whether or not there were victims with serious injuries and/or fatalities in the traffic accident with victims, we used a small sample with very imbalanced data. Empirical analysis supports the conclusion that, with a small sample of imbalanced data, machine-learning models generally do not perform better than statistical models; however, they perform similarly when the sample is large and has a small imbalance.
- Open Access
- 28 April 2022
- Published by MDPI AG in Information
- Vol. 13 (5) , 227
An Approach to Churn Prediction for Cloud Services Recommendation and User Retention
The digital world is very dynamic. The ability to timely identify possible vendor migration trends or customer loss risks is very important in cloud-based services. This work describes a churn risk prediction system and how it can be applied to guide cloud service providers for recommending adjustments in the service subscription level, both to promote rational resource consumption and to avoid CSP customer loss. A training dataset was built from real data about the customer, the subscribed service and its usage history, and it was used in a supervised machine-learning approach for prediction. Classification models were built and evaluated based on multilayer neural networks, AdaBoost and random forest algorithms. From the experiments with our dataset, the best results for a churn prediction were obtained with a random forest-based model, with 64 estimators, having 0.988 accuracy and 0.997 AUC value.
- 28 April 2022
- Published by MDPI AG in Information
- Vol. 13 (5) , 227
Open AccessAn Approach to Churn Prediction for Cloud Services Recommendation and User Retention
The digital world is very dynamic. The ability to timely identify possible vendor migration trends or customer loss risks is very important in cloud-based services. This work describes a churn risk prediction system and how it can be applied to guide cloud service providers for recommending adjustments in the service subscription level, both to promote rational resource consumption and to avoid CSP customer loss. A training dataset was built from real data about the customer, the subscribed service and its usage history, and it was used in a supervised machine-learning approach for prediction. Classification models were built and evaluated based on multilayer neural networks, AdaBoost and random forest algorithms. From the experiments with our dataset, the best results for a churn prediction were obtained with a random forest-based model, with 64 estimators, having 0.988 accuracy and 0.997 AUC value.
- Top Cited PapersOpen Access
Machine Learning Approaches to Traffic Accident Analysis and Hotspot Prediction
Traffic accidents are one of the most important concerns of the world, since they result in numerous casualties, injuries, and fatalities each year, as well as significant economic losses. There are many factors that are responsible for causing road accidents. If these factors can be better understood and predicted, it might be possible to take measures to mitigate the damages and its severity. The purpose of this work is to identify these factors using accident data from 2016 to 2019 from the district of Setúbal, Portugal. This work aims at developing models that can select a set of influential factors that may be used to classify the severity of an accident, supporting an analysis on the accident data. In addition, this study also proposes a predictive model for future road accidents based on past data. Various machine learning approaches are used to create these models. Supervised machine learning methods such as decision trees (DT), random forests (RF), logistic regression (LR), and naive Bayes (NB) are used, as well as unsupervised machine learning techniques including DBSCAN and hierarchical clustering. Results show that a rule-based model using the C5.0 algorithm is capable of accurately detecting the most relevant factors describing a road accident severity. Further, the results of the predictive model suggests the RF model could be a useful tool for forecasting accident hotspots.
Top Cited PapersOpen AccessMachine Learning Approaches to Traffic Accident Analysis and Hotspot Prediction
Traffic accidents are one of the most important concerns of the world, since they result in numerous casualties, injuries, and fatalities each year, as well as significant economic losses. There are many factors that are responsible for causing road accidents. If these factors can be better understood and predicted, it might be possible to take measures to mitigate the damages and its severity. The purpose of this work is to identify these factors using accident data from 2016 to 2019 from the district of Setúbal, Portugal. This work aims at developing models that can select a set of influential factors that may be used to classify the severity of an accident, supporting an analysis on the accident data. In addition, this study also proposes a predictive model for future road accidents based on past data. Various machine learning approaches are used to create these models. Supervised machine learning methods such as decision trees (DT), random forests (RF), logistic regression (LR), and naive Bayes (NB) are used, as well as unsupervised machine learning techniques including DBSCAN and hierarchical clustering. Results show that a rule-based model using the C5.0 algorithm is capable of accurately detecting the most relevant factors describing a road accident severity. Further, the results of the predictive model suggests the RF model could be a useful tool for forecasting accident hotspots.