Abstract
Mixed-type data is common in the real world. However, supervised learning algorithms such as support vector machines or neural networks can only process numerical features. One may choose to drop qualitative features, at the expense of possible loss of information. A better alternative is to encode them as new numerical features. Under the constraints of time, budget, and computational resources, we were motivated to search for a general-purpose encoder but found the existing benchmarks to be limited. We review these limitations and present an alternative. Our benchmark tests 16 encoding methods, on 15 regression datasets, using 7 distinct predictive models. The top general-purpose encoders were found to be Catboost, LeaveOneOut, and Target. © 2021, The Author(s), under exclusive license to Springer Nature Switzerland AG.

Abstract
Mobile robotics in forests is currently a hugely important topic due to the recurring appearance of forest wildfires. Thus, in-site management of forest inventory and biomass is required. To tackle this issue, this work presents a study on detection at the ground level of forest tree trunks in visible and thermal images using deep learning-based object detection methods. For this purpose, a forestry dataset composed of 2895 images was built and made publicly available. Using this dataset, five models were trained and benchmarked to detect the tree trunks. The selected models were SSD MobileNetV2, SSD Inception-v2, SSD ResNet50, SSDLite MobileDet and YOLOv4 Tiny. Promising results were obtained; for instance, YOLOv4 Tiny was the best model that achieved the highest AP (90%) and F1 score (89%). The inference time was also evaluated, for these models, on CPU and GPU. The results showed that YOLOv4 Tiny was the fastest detector running on GPU (8 ms). This work will enhance the development of vision perception systems for smarter forestry robots.

Abstract
Deepfake in multimedia content is being increasingly used in a plethora of cybercrimes, namely those related to digital kidnap, and ransomware. Criminal investigation has been challenged in detecting manipulated multimedia material, by applying machine learning techniques to distinguish between fake and genuine photos and videos. This paper aims to present a Support Vector Machines (SVM) based method to detect tampered photos. The method was implemented in Python and integrated as a new module in the widely used digital forensics application Autopsy. The method processes a set of features resulting from the application of a Discrete Fourier Transform (DFT) in each photo. The experiments were made in a new and large dataset of classified photos containing both legitimate and manipulated photos, and composed of objects and faces. The results obtained were promising and reveal the appropriateness of using this method embedded in Autopsy, to help in criminal investigation activities and digital forensics.

Abstract
Experienciar a cidade enquanto espaço de aprendizagem implica entrelaçamento de seus vários tempos, espaços e de suas dimensões humanas e não humanas, digitais, biológicas, históricas, econômicas, etc. Entender a cidade como entidade viva, complexa e comunicativa se afasta da visão antropocêntrica de mundo e instaura outras formas de comunicação, habitação e aprendizagem. Portanto, este artigo apresenta o processo de desenvolvimento de uma prática pedagógica gamificada cujo objetivo é compreender como o coengendramento dos diversos entes constituintes da cidade produz o pensamento computacional na perspectiva da aprendizagem inventiva, segundo uma proposta de Educação OnLIFE. Tal prática integra o projeto de pesquisa A cidade como espac¸o de aprendizagem: pra´ticas pedago´gicas inovadoras para a promoc¸a~o da cidadania e do desenvolvimento social sustenta´vel, financiado pela Fundação Carlos Chagas e pelo Itaú Social, e desenvolvida pelo Grupo de Pesquisa Educação Digital — GPe-dU, UNISINOS/CNPq. O estudo, qualitativo, apropria-se do método cartográfico de pesquisa-intervenção para produção e análise dos dados. Como resultado, apresentam-se pistas que permitem compreender a emergência do pensamento computacional para explorar a cidade, bem como a necessidade de inovação de práticas pedagógicas a partir da problematização do mundo para uma Educação conectada com a vida.

Abstract
In democratic countries, government websites became an important channel for interaction with the public administration in the last few years. Nevertheless, several issues have an impact on the way users access to content and information. Lack of accessibility and usability or, in the broad sense, lack of concern with user needs, can still be found in many government websites. To address the problem, a previous literature review on e-government information delivery attributes was performed. Based on this review, a large set of attributes related to quality was obtained to evaluate these dimensions in the context of e-government. The purpose of this study is to better understand which of these attributes are the most valued, in the users' perspective, for evaluating content delivered by government websites. A qualitative approach was adopted, using Focus Group interviews as a strategy to obtain data and Thematic Analysis to analyze such data. The main results highlighted the attributes related to content delivery, interaction, and emotional aspects. User Experience, accessibility, and usability were prioritized by Focus Group participants.

Abstract
Nowadays, as more data is now available from an increasing number of installed sensors, load forecasting applied to buildings is being increasingly explored. The amount and quality of resulting information can provide inputs for smarter decisions when managing and operating office buildings. In this article, the authors use two data-driven methods (artificial neural networks and support vector machines) to predict the heating and cooling energy demand in an office building located in Lisbon, Portugal. In the present case-study, these methods prove to be an accurate and appealing alternative to the use of accurate but time-consuming multi-zone dynamic simulation tools, which strongly depend on several parameters to be inserted and user expertise to calibrate the model. Artificial neural networks and support vector machines were developed and parametrized using historical data and different sets of exogenous variables to encounter the best performance combinations for both the heating and cooling periods of a year. In the case of support vector regression, a variation introduced simulated annealing to guide the search for different combinations of hyperparameters. After a feature selection stage for each individual method, the results for the different methods were compared, based on error metrics and distributions. The outputs of the study include the most suitable methodology for each season, and also the features (historical load records, but also exogenous features such as outdoor temperature, relative humidity or occupancy profile) that led to the most accurate models. Results clearly show there is a potential for faster, yet accurate machine-learning based forecasting methods to replace well-established, very accurate but time-consuming multi-zone dynamic simulation tools to forecast building energy consumption.