With the widespread use of blended English teaching, the teaching model has become increasingly rigid and needs to be reformed. This paper proposes an analytical model of student learning behavior under blended English teaching based on big spatiotemporal data and constructs a structure of offline and online student learning behavior. The DWT-FCM clustering method for learning behaviors is constructed by using a similar distance based on time series instead of Euclidean distance in the traditional fuzzy C-mean algorithm. The maximum information coefficient was used to analyze the correlation between learning behaviors and grades based on MI. the DWT-FCM algorithm divided learning behaviors into five clusters, where the discussion number of cluster 0 reached an average of 45.2, which was 3.2 times higher than that of cluster 1. Cluster 4 has an average number of task completions of only 1.88, which will skip almost the entire course taken. The correlations of the number of task completion, course video progress, and assignment evaluation scores are all highly correlated, with correlation coefficients reaching 0.884 and 0.825, respectively. This study suggests the influence of learning characteristics on English performance in a blended teaching model, which can effectively guide the reform of the blended English teaching model.