Learning Data Mining and Learning Effect Prediction under Multi-mode: A case study of C language course

In this paper, a classification model of teaching effect based on long short-term memory network (LSTM) is proposed based on intelligent evaluation of C language teaching effect. Aiming at the multi-dimensional behavior and performance characteristics in the process of language learning, the weighted fusion method integrates 14 key indicators, such as code cyclomatic complexity, code error rate, video learning duration, course visit frequency, regular exam scores, final scores, etc., to construct a time series feature vector reflecting the comprehensive learning state of students. Based on the time-dependent learning behavior of LSTM network modeling, the four-level classification results of "excellent, good, medium and poor" are output. To verify the effectiveness of feature fusion, a comparative experimental group was designed: one experiment used all 14 weighted features as inputs, and the other experiments used features of a single class or fewer classes. The experimental results show that the classification accuracy of the all-feature fusion model reaches 88%, which is significantly higher than that of using only a single pattern or few patterns. This study confirms the necessity of multi-modal feature fusion in time series teaching evaluation, and provides data-driven decision support for dynamic analysis of language course teaching effect and personalized intervention.