Abstract
More and more technological advances offer new paradigms for training, allowing novel forms of teaching and learning to be devised. A widely accepted prediction is that computing will move to the background, weaving itself into the fabric of our everyday living spaces and projecting the human user into the foreground. This forecast turns out to be an opportunity for human-computer interaction as a way to monitor and assess the user’s stress levels during high-risk tasks. The main effects of stress are increased physiological arousal, somatic complaints, mood disturbances (anxiety, fear and anger) and diminished quality of working life (e.g. reduced job satisfaction). To mitigate these problems, it is necessary to detect stressful users and apply coping measures to manage stress. Human-computer interaction could be improved by having machines naturally monitor their users’ stress, in a non-invasive and non-intrusive way. This article discusses the development of a random forest classifier with the goal of enabling the assessment of high school students’ stress during academic exams, through the analysis of mouse behaviour and decision-making patterns. © Springer Nature Switzerland AG 2019.

Abstract
As the field of Machine Learning evolves, the number of available learning algorithms and their parameters continues to grow. On the one hand, this is positive as it allows for the finding of potentially more accurate models. On the other hand, however, it also makes the process of finding the right model more complex, given the number of possible configurations. Traditionally, data scientists rely on trial-and-error or brute force procedures, which are costly, or on their own intuition or expertise, which is hard to acquire. In this paper we propose an approach for algorithm recommendation based on meta-learning. The approach can be used in real-time to predict the best n algorithms (based on a selected performance metric) and their configuration, for a given ML problem. We evaluate it through cross-validation, and by comparing it against an Auto ML approach, in terms of accuracy and time. Results show that the proposed approach recommends algorithms that are similar to those of traditional approaches, in terms of performance, in just a fraction of the time.

Abstract
Demographic changes are leading to a growing older population (>65 years), with repercussions on age-related conditions. From a Computer Science perspective, this also means that there will soon be a significant number of users with changes in perceptual and motor skill capacities. The goal of this work is to develop an environment to support the preservation of memory and functional capacities of the elderly. Health professionals will be able to set up and personalize immersive and realistic scenarios with high ecological validity composed of visual, auditory, and physical stimuli. Patients will navigate through and interact with these scenarios and stimulate memory functions by later recalling distinct aspects of the different exercises of the tool. The long-term goal is to build a behavioral model of how older users interact with technology. © Springer Nature Switzerland AG 2019.

Abstract
Nowadays, it is common for higher education institutions to use computer-based exams, partly or integrally, in their evaluation processes. These exams, much like their paper-based counterparts, are one of the most significant sources of stress in the life of students. However, the fact that exams are undertaken in a computer allows for new features to be acquired that may provide more reliable insights into the behaviour and state of the student during the exam. In this article we analyse these novel behavioural features and explore, to which extent, they can point out previously unknown phenomena. Specifically, we show that the time a student takes to complete an exam is correlated with mouse dynamics features. In practical terms, we are able to predict the duration of each individual exam with a satisfying error based on the interaction patterns of the student.

Abstract
In the last years, organizations and companies in general have found the true potential value of collecting and using data for supporting decision-making. As a consequence, data are being collected at an unprecedented rate. This poses several challenges, including, for example, regarding the storage and processing of these data. Machine Learning (ML) is also not an exception, in the sense that algorithms must now deal with novel challenges, such as learn from streaming data or deal with concept drift. ML engineers also have a harder task when it comes to selecting the most appropriate model, given the wealth of algorithms and possible configurations that exist nowadays. At the same time, training time is a stronger restriction as the computational complexity of the training model increases. In this paper we propose a framework for dealing with these challenges, based on meta-learning. Specifically, we tackle two well-defined problems: automatic algorithm selection and continuous algorithm updates that do not require the retraining of the whole algorithm to adapt to new data. Results show that the proposed framework can contribute to ameliorate the identified issues.

Abstract
In the last years, the number of machine learning algorithms and their parameters has increased significantly. On the one hand, this increases the chances of finding better models. On the other hand, it increases the complexity of the task of training a model, as the search space expands significantly. As the size of datasets also grows, traditional approaches based on extensive search start to become prohibitively expensive in terms of computational resources and time, especially in data streaming scenarios. This paper describes an approach based on meta-learning that tackles two main challenges. The first is to predict key performance indicators of machine learning models. The second is to recommend the best algorithm/configuration for training a model for a given machine learning problem. When compared to a state-of-the-art method (AutoML), the proposed approach is up to 130x faster and only 4% worse in terms of average model quality. Hence, it is especially suited for scenarios in which models need to be updated regularly, such as in streaming scenarios with big data, in which some accuracy can be traded for a much shorter model training time.