In the Corona Endemic, we are not just returning to offline education patterns but are already moving towards education 5.0. Online, normal, blended learning patterns have become commonplace. Online learning assessment requires fast and precise predictions of student performance (high accuracy). The reason is first, due to limited direct interaction. Second, normal learning usually involves an assessment of the learning process and character assessment to be able to provide an accurate final assessment, which is difficult to implement in online learning accurately. Third, there is a lot of data to be processed quickly and precisely so that it can be reported to educational institutions and to students' families. Fourth, Informatics is a lesson that is 80% practical and 20% theory so that the assessment instruments used are 80% performance instruments (Bloom's taxonomy: C2, C3, C4, C5) and 20% multiple choice instruments (C1). Informatics correction and assessment requires more time because 80% cannot be assessed automatically. This research aims to predict student performance (Pass (1) or Intervention (0)) on the results of the online learning process assessment for informatics subjects in high school. If the student performance prediction results in an intervention, it will be immediately followed up by providing an intervention strategy to increase student performance. The target of the research results is to achieve > 70% accuracy on the processed dataset. This research uses the ensemble learning method random Forest Classification and XG Boosting classification. The research results of Student Performance Prediction using XG Boost Classification produce higher accuracy than RF Classification which has an average accuracy value = 93% while RF Classification has an average accuracy result = 92%. The research objectives have been achieved because the results of the 2 methods used have met the desired targets.