Clustering of college students based on improved K-means algorithm

Many colleges have accumulated a large amount of information, such as achievement data and consumption records. According to the above information, we attempt to identify the student group from various aspects. Based on this, we can acquire the characteristics of students in different groups, then get the relationship between students’ different behaviors by association rules mining method. In this way, the college can have a better understanding of students to accomplish the reasonable management. In order to obtain more accurate cluster results, we proposed an improved K-means algorithm. Specially, we effectively detect outliers based on the grid density. In addition, we design a new method to produce initial cluster centers which replaces the traditional random way. Real experiments are conducted and the results show the iteration time is reduced and clustering precision is improved.