There is a consensus that serious games have a significant potential as a tool for instruction. However, their effectiveness in terms of learning outcomes is still understudied mainly due to the complexity involved in assessing intangible measures. A systematic approach—based on established principles and guidelines—is necessary to enhance the design of serious games, and many studies lack a rigorous assessment. An important aspect in the evaluation of serious games, like other educational tools, is user performance assessment. This is an important area of exploration because serious games are intended to evaluate the learning progress as well as the outcomes. This also emphasizes the importance of providing appropriate feedback to the player. Moreover, performance assessment enables adaptivity and personalization to meet individual needs in various aspects, such as learning styles, information provision rates, feedback, and so forth. This paper first reviews related literature regarding the educational effectiveness of serious games. It then discusses how to assess the learning impact of serious games and methods for competence and skill assessment. Finally, it suggests two major directions for future research: characterization of the player’s activity and better integration of assessment in games.

Serious Games represent an acknowledged potential for instruction, because they are able to strongly motivate learners. They can also provide immersive environments where advanced users can practice knowledge and skills, also exploiting multimodal interaction. They can combine the effectiveness of computer processing and data storage, with high levels of attractiveness. Our work has investigated the state of the art research on SGs, starting from the cognitive aspects, that are necessary in order to root technological development and applications in sound theoretical foundations. The paper discusses some key aspects about SG design and exploitation: choice of components-off-the-shelf or from-scratch design, tools and methodologies for development or adaptation, intelligent tutoring, virtual coaches and affective learning, living worlds, game mechanics, Human-Computer Interaction. While several SGs have been developed, still the literature stresses a lack of significant, extensive user tests. Further research is necessary to investigate in greater detail the real effectiveness of the various types of SGs. The paper proposes several research questions - that range from requirements elicitation to design and from deployment to use and evaluation - to be answered in order to avoid technology pushing and drive technological research according to the requirements of the end-users and stakeholders. We believe that deepening the analysis about these issues is key to strengthen the foundations of SG research, for which we identify four major directions: definition of metrics and learning progress evaluation tools; methodologies and tools for designing games from various topics and for various users; computing and communication architectures; technologies that can enhance the overall system performance.

This paper presents the Edge Learning Machine (ELM), a machine learning framework for edge devices, which manages the training phase on a desktop computer and performs inferences on microcontrollers. The framework implements, in a platform-independent C language, three supervised machine learning algorithms (Support Vector Machine (SVM) with a linear kernel, k-Nearest Neighbors (K-NN), and Decision Tree (DT)), and exploits STM X-Cube-AI to implement Artificial Neural Networks (ANNs) on STM32 Nucleo boards. We investigated the performance of these algorithms on six embedded boards and six datasets (four classifications and two regression). Our analysis—which aims to plug a gap in the literature—shows that the target platforms allow us to achieve the same performance score as a desktop machine, with a similar time latency. ANN performs better than the other algorithms in most cases, with no difference among the target devices. We observed that increasing the depth of an NN improves performance, up to a saturation level. k-NN performs similarly to ANN and, in one case, even better, but requires all the training sets to be kept in the inference phase, posing a significant memory demand, which can be afforded only by high-end edge devices. DT performance has a larger variance across datasets. In general, several factors impact performance in different ways across datasets. This highlights the importance of a framework like ELM, which is able to train and compare different algorithms. To support the developer community, ELM is released on an open-source basis.