Abstract: To deal with the problem of uneven data distribution and the situation that many factors easily cause inaccurate prediction, this paper proposes an accurate regression prediction method combining coarse to fine and feature selection. First of all, this paper puts forward a prediction method from coarse-to-fine, which uses decision trees to roughly classify the sub-intervals of the prediction target for overcoming the problem that direct prediction may be difficult to accurately fit due to uneven data distribution and large prediction intervals. Next, taking into account of the small amount of data and the circumstance that many feature factors may cause over-fitting and some redundant features may affect the prediction accuracy of the model, this paper proposes a regression prediction method based on feature selection to improve the prediction accuracy. Compared with the traditional regression model, the experimental results on the data set of English achievement of college students and their personality factors have proved that the method from coarse-to-fine and feature selection proposed in this paper has higher precision and better stability. More flexible and feasible syllabus based on proposed method could be laid out to tackle students’ personality traits problem to enhance their competition capability and thus impel English education in China in the future.

Key words: decision trees, coarse-to-fine, feature selection, regression prediction method