Educational data mining is currently a prominent topic. Existing methods such as graph convolutional networks often struggle with high-dimensional complex features and do not effectively utilize the weights and influences between features. To assist teachers in evaluating massive open online courses (MOOCs) and providing timely academic warnings for students. First, we extracted a number of features. Second, we design various feature extraction methods to analyze data from the C programming class on MOOCs. Third, we developed a predictive model incorporating sequence convolution and an improved attention mechanism to enhance the processing of time series features and feature weights. The results indicate that our method outperforms traditional models in terms of prediction accuracy.