Through the growing trend of online courses and digital learning tools, quantitative educational research is steadily evolving. The field of learning analytics, as it is called, is especially suitable for computer science education, as many programming environments are capable of logging fine-grained user data. Learning to program is often considered a difficult task, but those who persevere through the challenge may serve as a guiding light in improving computer science education. As such, the goal of this thesis is to identify those who have improved the most in an introductory programming course using learning analytics. For this, two methods were explored: analysing programming behaviour using error quotient, a previously established metric within learning analytics in computer science education, and analysing the residuals of a machine learning model trained on accumulative student activity throughout different periods of the course. Both methods showed clear distinction of students, and both provided different sets of students who stood out from the rest in their progression. However, further analysis and interviews with the teacher and students themselves are necessary to confirm the validity of the results.