Prediction of student dropout is a major challenge for educational institutions. Although AI-based approaches can predict dropout through data analysis, they often rely on a static set of features and fail to adapt to the evolving nature of data over time, thus limiting performance. Other application domains have the same characteristic of evolving data and several algorithms have been proposed to ensure dynamic feature selection over time. However, they are not adapted to dropout prediction, characterized by imbalanced data, which limits their effectiveness. In this paper we introduce CI-DFS, a novel algorithm that achieves dynamic feature selection on class imbalanced data and used to perform dropout prediction. CI-DFS relies on three main elements: 1) a weighted mutual information-based metric to deal with imbalanced class distributions, 2) an adaptive threshold mechanism to dynamically assess feature relevance, and 3) a temporal drift management system to ensure feature relevance over time. CI-DFS has been evaluated on two real-world benchmark educational datasets, and compared with a state-of-the-art dynamic feature selection algorithm. CI-DFS outperforms this algorithm in three key aspects: it improves dropout prediction by 9%, reduces the number of features selected by over 50%, and significantly accelerates feature selection time, making it 10 times faster.

International audience

The use of Artificial Intelligence (AI), notably Machine Learning (ML), is gaining momentum in archaeology, opening up new possibilities such as artifact classification, site location prediction, and remains analysis. One of the major challenges in this regard is the lack of qualified archaeologists who are experts in machine learning. In this study, we introduce IArch, a tool that enables eXplainable Artificial Intelligence (XAI) data analytics for archaeologists without requiring specific programming skills. It specifically allows data analysis performed to either validate existing data-supported hypotheses or generate new ones. The tool covers the entire workflow for applying ML, from data processing to explaining the final results. The tool allows the use of supervised and unsupervised ML algorithms, as well as the SHapley Additive exPlanations (SHAP) technique to provide archaeologists with global and individual explanations for the predictions. We demonstrate its use on data from a Xiongnu cemetery (100 BC/AD 100) in the Mongolian steppes.

International audience

To fulfill its complex business demands, an organization may require the integration of many SaaS applications. A main challenge in this regard is to select the suitable SaaS offerings that satisfy all the business requirements. Due to the diversity of QoS needs and the large number of SaaS providers, selecting the suitable set of SaaS applications to be integrated may result in QoS degradation and higher costs. In this paper, we propose to use a System-of-Systems (SoS)-based approach for the selection and integration of SaaS. We believe that the use of SoS and its dynamic properties could improve the service selection at SaaS level. To do so, we propose three Mixed Integer Programming models that optimize the global SaaS behavior by minimizing the overall delivery costs, and maximizing the global functional and QoS utilities. The implementation of the approach and the experimental results highlight the efficiency of our proposed solution.

Scalability is a key issue for Multi-Agent Systems (MAS) that aim to model and simulate complex systems. Distributed infrastructures such as clusters, grids and clouds are powerful computational environments that can be effectively used to run large-scale agent-based simulations. To properly distribute an agent-based system and ensure its performance, an appropriate partitioning approach is required. Although multiple partition methods for distributed MAS exist, they remain specific to the individual requirements of a given application domain. There is no generic approach for guiding the designers and developers to select an appropriate approach for partitioning a given agent-based system. Thus a recurrent challenging task, for MAS designers and developers, is how to evaluate, select and then apply the appropriate partitioning mechanism for a given MAS. Therefore, in this paper, we present a generic conceptual framework useful to analyze existing partitioning methods. It can also be used as a basis while designing a distributed architecture of new MAS.