Accurate language achievement prediction method based on multi-model ensemble using personality factors

To overcome the noise in personality factors and precisely predict language achievement, we propose a robust regression algorithm based on the maximum correntropy criterion (MCC) and the coarse-to-fine method. Firstly, as there are many samples while few personality factors correlate to the language achievement in the data set, we propose a regression method based on Pearson feature selection to eliminate the noise and redundant features for solving the overfitting problem. Secondly, as the learning ability of each traditional regression model is different and limited, we introduce the model ensemble method based on MCC to predict language achievement via personality factors. Thirdly, owing to the fact that the language achievement data is unevenly distributed and the same model parameter cannot fit all the data effectively, we propose a coarse-to-fine prediction method to reduce prediction errors, which divides the range of the language achievement into multiple intervals and then establishes different regression models at each interval to obtain more accurate results. The experimental results on the data set of the personality factors and English achievement demonstrate the high precision and robustness of the proposed algorithm compared with the traditional single regression models.