In the realm of digital education, the growing use of short-form online videos, coupled with innovative generative AI methods, has dramatically expanded the production of didactic academic videos. This shift, however, underscores a critical question - how to ascertain the "effectiveness" of these videos for student learning? It is essential to devise a classification mechanism that filters videos for clarity, comprehensibility, and their capacity to meet student learning objectives. The automated evaluation of these learning videos holds substantial implications for student academic performance. Accordingly, this paper presents a novel supervised-learning-based approach, predicated on video feature analysis, to predict the effectiveness of K-12 science and mathematics videos. Our method integrates diverse features such as image, spoken text, and audio, among other hand-crafted elements, to accurately assess video effectiveness. We conduct an evaluation of our approach using a comprehensive dataset comprised of 3,134 short-form academic videos. The results demonstrate robust performance, with the system achieving an accuracy of 76.1% and an F1 score of 80.6%.