Multimodal learning analytics in educational data mining is an increasingly popular research field aimed at improving student learning outcomes by applying techniques to a diverse range of data sources. This study proposes a multimodal learning analytics framework using machine learning for early student performance prediction, leveraging data from various sources. Such early-stage insights provide actionable intelligence, allowing educators to implement timely interventions and personalized support strategies to address challenges proactively. Our proposed model classified student performance into four distinct outcomes: Distinction, Pass, Fail, and Withdraw and was validated using the Open University Learning Analytics Dataset (OULAD), achieving a notable accuracy of 81.67% and an F1-score of 81.40% for multi-class classification. For binary classification tasks, the model attained an impressive accuracy of 97.45%. This study demonstrates how predictive analytics can be used to improve teaching methods, encourage proactive educational interventions, and raise overall student success rates by focusing on multimodal data integration and early prediction.