With the booming development of large-scale open online courses, blended teaching, which combines traditional closed teaching and online open teaching, is increasingly favored by colleges and universities. In this paper, from the perspective of blended teaching of English in colleges and universities, based on the LSTM model to predict the relevant learning data in English teaching in colleges and universities, and based on the density optimization K-mean algorithm to cluster the student subjects with different learning behaviors, and then use the Apriori algorithm to study the correlation rules of the learning effectiveness and behaviors, to provide ideas for English teaching in colleges and universities. The clustering results show that the average learning scores of the first, second and third categories of learners are 92.35, 83.57 and 64.96 respectively. The results of association rule analysis show that routinely, the more active learners are in each learning session, the greater the possibility of getting better learning outcomes. The LSTM learning prediction model Precision, Recall and F1 assessment indexes trained with 4-month behavioral data are 0.899, 0.785 and 0.833 respectively, which are all greater than the corresponding index values of SVM, MLP and RF models, and have a significant advantage in prediction effect. This study provides lessons and references for improving the effectiveness of English teaching in colleges and universities.