Maximizing learning for all students in digital learning environments could be achieved by using real-time personalized support for Self-Regulated Learning (SRL), a critical skill where learners monitor and guide their learning. To provide such SRL support, the ability to track students’ SRL during learning is key. While SRL has been extensively studied, measuring SRL through the evolving text remains largely unexplored. Yet, such real-time measurement could enable real-time support during writing. Real-time performance tracking transforms personalized support from reactive to proactive, ensuring that students get the help they need exactly when needed, ultimately enhancing students’ learning outcomes and SRL skills. Therefore, this study explores automated SRL detection in evolving essay texts using natural language processing (NLP) methods. Multiple Machine Learning (ML) models were trained and evaluated to assess their effectiveness in classifying SRL processes during writing. Specifically, this research addresses the question: To what extent can different ML models accurately classify different cognitive and metacognitive processes students apply during writing?. Keystroke data was feature-engineered into time-based segments to facilitate real-time application, capturing the evolving essay text. The model’s classifications were compared against manual coding to assess the reliability and accuracy. The model’s results show an average accuracy of 0.78, Cohen’s Kappa of 0.62, and a macro F1-score of 0.45. These results indicate that NLP methods can be used to classify SRL to inform real-time personalized support.