Abstract: Learners always have their learning style preferences according to their different cognitive processes. Auto- matically modeling the learners' learning style can get the more accurate information compared with the questionnaires which is free from the problem of inaccurate self-conceptions of the learners. There are many problems in the current LS detecting methods, like only can detect the I_S dimensions in a specific model, can not adaptively adjust the I_S prefer- ences in a different learning environment. We provided a new method to forecast the learners' LSs which is called LS- Pre. In LS-Pre, non-linear dynamic programming is used to construct the mathematic model while simulated annealing algorithm is used to optimize the goal function. We illustrated the effectiveness of I_S-Pre in part 4 of this paper.

Key words: Learning style(LS) , Learning style dimension, Behavior matrix