Personalized feedback has been proven to be an effective mechanism for supporting learners during the learning process. Receiving feedback before and after the elaboration of the assignments improves learners' performance and engagement. However, the major drawback when providing personalized feedback is the teachers' effort due to the need for more scalability.

Information and Communication Technologies (ICT) have some potential to fully or partially automate the process. For instance, learning activities can be adapted as quizzes with automated assessment but limiting the provided feedback, generative artificial intelligence tools can be used to give feedback based on rubric assessment or by assessing the learners’ activity. In this case, feedback is richer, but the teacher needs to revise the provided feedback. Therefore, while technology helps, quality feedback (i.e. feedback that contributes to learning) relies on the teacher's effort, knowledge of the learners’ progression, and the implementation of specific strategies to ensure these learners’ progression (formative assessment).

Knowledge about learners' progression can be obtained by using predictive analytics. Learning management systems (LMS) produce a massive amount of data every day during the learners' learning process. The learners' interaction within these systems generates a digital trace composed of data, such as navigational data, textual data, the patterns learners follow when they access the available learning resources, etc. Analysing these rich data sets implies knowing better, finding out their profile, or even identifying failure or dropout at-risk situations that may jeopardize their success.

This paper explores strategies to enhance the feedback provided to the learners. Different scenarios are proposed, including different feedback strategies with dropout and failure at-risk information to expect an impact on learners' retention. To support teacher effectiveness, the predictive analytics system LIS (Learning Intelligent System), currently under development at Universitat Oberta de Catalunya (UOC), provides at-risk identification and feedback recommendations.

This paper's contribution is based on the design strategies to provide personalized feedback through the LIS system, which will help teachers when giving feedback, but mainly learners, succeed in their online courses. By predicting their failure and dropout risk using data from the CANVAS LMS and their performance level, the system will recommend feedback to teachers to help and encourage learners to achieve their learning goals and, consequently, their course certification. The findings will contribute to improving the teacher-learning process at UOC and also to other blended and online environments.

Keywords: Personalized feedback, learning management system, predictive analytics, learning intelligent system, learning analytics, e-assessment, e-learning, feedback, online education.