基于海量数据挖掘的知识学习效果预测方法分析-analysis of knowledge learning effect prediction method based on massive data mining.docx

AbstractAs human society entered the information age, information technology has made a great effect on the education and become the power and technology foundation for the education reform. Educational data mining has become a newly emerging study field. Education informatization development makes all kinds of data in educational fields grow fast. How to get useful information from the mass of data for educators and learners to improve education management performance and learning performance is the reason why educational data mining research appears.This paper is focus on how to get useful information from mass educational data for learners and educational decision-making departments and make an analysis on the mass data learning effects data mining. The system is to predict students’ future examination performance according to their historical behaviors and ultimately improve the students’ study process and adjust the teaching methods immediately. The system includes two parts: the first part is the feature generation and the second part is about the predition algorithm.The first part is about feature generation. First take an analysis on the data. First is the validation set generation. The validation set generation is based on the temporal information. Because there are many contents of the records between students and the tutor systems interaction, so according to the meaning of features, select the most useful features to ensure the accuracy and feature files size. Also, some complex features need further processed, for example, feature binarization and normalization. Temporal information features is extracted from the original dataset. To better display the features, features are scaled between zero and one by some functions and extended into binary features.The second part is about the prediction. To achieve better experimental results, this paper tries to use two kinds of classifier. The first one is K-nearest algorithm, and we improve the algorithm in two