Fuzzy Logic-Based Preliminary Risk Assessment for a Supervised Federated Learning Solution in Predicting Student Results

Accurate prediction of student results plays a key role in academic performance improvement and definition of personalized learning paths. A fuzzy logic based model can easily incorporate uncertainties and vagueness that are inherently present in this task. The data necessary for training of such a model is sensitive especially when they are collected from more than one university. Federated learning provides a framework for the seamless cooperation of universities in the learning process of a model. The process is decentralized, each institution trains a model with its own data while the final model is elaborated by aggregating the local models. This approach ensures data privacy and security within the limits given by data access rights. In this paper, a preliminary risk assessment is presented that is carried out using the case of a federated learning project of two universities aiming the development of a fuzzy logic-based student result prediction model. The discussed risk analysis encompasses several aspects including risks related to technical issues, model accuracy, model performance, data security, challenges of the implementation, etc. Following the fuzzy rating of individual aspects, a fuzzy inference-based aggregation is carried out followed by a defuzzification to determine a final risk value. Based on it, risks are prioritized and countermeasures are defined to alleviate them. The presented risk evaluation case study provides valuable insight into the potential threats of the investigated implementation of federated learning, while also highlighting the potential in this technology.