This study takes college students’ English learning behavior as an entry point and proposes a learning behavior analysis model based on multilayer perceptron (MLP). Through the synergy between the interactive feature capture module and the basic feature capture module, the accurate modeling of higher-order features such as learning concentration and initiative is realized. The English learning behavior data of 218 students were collected through the TULIP Smart Learning Platform, and a four-dimensional indicator system containing learners’ basic information, operational behavior, collaborative behavior and problem solving behavior was constructed. In the teaching experiment, the designed MLP model realizes the accurate prediction of key behaviors such as learning concentration and initiative through the interactive feature capture module and the basic feature capture module. More than 90% of students chose to agree or strongly agree in the survey results of each test item of the modelassisted learning dimension. More than 90% of students believe that remedial learning is able to target their weak points, reduce their study load and improve their learning outcomes. More than 90% of the students said that the personalized teaching strategy is compatible with their own learning habits, and the average score of the personalized teaching strategy dimension is 4.63, which proves that using the proposed model to analyze learning behaviors and formulate targeted personalized teaching strategies can effectively improve learning effectiveness.
