This study examines the effectiveness of machine learning models in predicting student performance in a Physics-I course using a blended learning approach. The dataset included interactions from 195 students with educational videos, online quizzes, and face-to-face final exam results. The course was structured with interactive video lectures and online quizzes during the first four weeks, followed by traditional in-class instruction for the remaining five weeks. To predict student performance, machine learning methods such as k-nearest Neighbors, Random Forest, kNN, and Logistic Regression were utilized. The k-nearest Neighbors model achieved the highest prediction accuracy at 86%, effectively categorizing students into low and high achievers. Online quiz scores and interactions with interactive video lectures were identified as the most significant predictors of performance. The findings underscore the potential of machine learning to create personalized learning experiences, provide early interventions, and improve course design.