以混合式分群演算法探討學生學習成效之研究

大數據資料探勘時代下，有效掌握學生學習成效成了近年校務研究的重要議題。本研究將校務研究中心提供的學生四年成績資料加以串接整理，並嘗試多種分群法後整理出P-H分群法，進而從中觀察學習成效。P-H分群融合分割式分群與階層式分群，從歐式距離與餘弦距離的不同定義，由成績在四年間的走向區分出不同特性族群間學生的差異，據以進行統計與關聯規則分析。本研究在視覺呈現上除了基本統計圖表亦使用Embedding Projector工具將資料做PCA降維，並能直觀地選取分群點的鄰居，以理解資料分佈之特性與性質。研究結果確立了不同群間學期成績的差異，並以社團與陸生為例顯示其在群間關聯規則的影響性。

關鍵字

校務研究；學習成效分析；機器學習； Scikit-learn套件； Embedding Projector ；關聯規則

並列摘要

Under the era of big data exploration, effective mastery of student learning has become an important topic in institutional research in recent years. In this study, the four-year student data provided by the Institutional Research Center are concatenated and attempting a variety of clustering methods, a hybrid clustering method, called P-H Clustering, is established The learning effectiveness can be observed from the P-H Clustering. The P-H Clustering adopts the combination of partition-based clustering and hierarchical clustering, based on the different definitions of Euclidean distance and cosine distance. It effectively separates the different patterns among students in different clusters for applying statistical analysis and association rules. In this study, in addition to the basic statistical charts, the Embedding Projector tool is used to reduce the dimension of the data in PCA and select the neighbor points of the cluster in visualization. The results of the study identified differences in semester grades between different clusters. It also showed the impact of associations rules between club students and mainland China's students.