Classification演算法的特色是分成兩個階段,第一個階段是training,用已經分類的資料並根據資料的特徵做出對應的類別,第二個階段是Classification,對其他未經分類資料的特徵做分類。DGAGC是一種Classification演算法,適用於離散型資料,連續型資料需要額外處理。我們過去的研究已經讓DGAGC支援Hadoop MapReduce運算模型。但是Hadoop MapReduce的版本只針對DGAGC training的部分。在Classification部分,只有單機版本。其中以training的部分最花時間。本篇論文提出了Spark版本的DGAGC training與Classification,藉此來改善Hadoop版本在資料集運算量不算大時的執行效率。再來是DGAGC Classification的部分,單機版本在預測模型太大的時候就無法進去預測。所以提出Spark版本的DGAGC Classification改善此問題。;The DGAGC algorithm, developed by National Central University, is a classification algorithm based on association-rule mining and searching. The DGAGC algorithm also specifies a distributed computing approach for model training, which is implemented on top of Hadoop MapReduce. In this study, we propose a new distributed computing approach for the DGAGC algorithm based on Apache Spark. With the support of in-memory computing by Spark, the new distributed DGAGC algorithm can achieve less average execution time for model training, given four different training data sets. In addition, we also propose a distributed version of the DGAGC for data classification.