There are no teachers to evaluate the learning state of the learners in the online learning system, so most of the learners need to set their own learning strategies and adjust their learning steps, which lead to the lack of individualized guidance for the learners and the poor efficiency of the learners. In order to solve this problem, a diagnostic evaluation model was proposed in the College English Diagnostic Test System. The model analyzed the data collected in the database from the four aspects of learners' learning condition, question types association analysis, knowledge point association analysis, and CET-4 grade prediction, and it used Student-Problem Chart analysis method, data mining, and machine learning to build models respectively. Finally, the diagnostic evaluation model was obtained by merging these four sub models and it was implanted to improve the College English Diagnostic Test System. The experimental results showed that the diagnostic evaluation can effectively help learners to practice and improve their grades of CET-4.