資訊管理學報
葉怡成;杜進明;丁導民;王逸芸;劉謹豪;
頁: 51-78
日期: 2008/10
摘要: 本研究所提出的關聯推理神經網路(Association Reasoning Neural Networks, ARNN)是修改自倒傳遞神經網路演算法，可以產生關聯規則，為傳統的關聯分析開啟完全不同的研究途徑。經由一個數值例題與一個實際例題的結果歸納出以下結論：(1)ARNN的推論輸出值與資料中的信賴度大約相等。(2)當ARNN的隱藏單元數減少時，檢出率、準確率會跟著變小。(3)ARNN可以利用減少隱藏單元數來避免產生信賴度過低的關聯規則。(4)ARNN產生的規則與由傳統關聯分析法所找到的規則重疊性很高，且能找出一些被傳統關聯分析法忽略的關聯規則。
關鍵字: 資料探勘;類神經網路;關聯規則;關聯分析;
Abstract: In this study, we proposed the Association Reasoning Neural Network (ARNN) which is derived from ANN, can produce association rules, and open a new approach for association analysis. Based on a numerical and a practical example, some conclusions can be gotten: (1) The reasoning output value of ARNN is proportional to the confidence of the association rules hidden in the data set. (2) As the number of neurons in the hidden layer of ARNN is decreasing, the detection rate and the accuracy rate are decreasing. (3) The ARNN can avoid producing the association rules with low support value by reducing the number of neurons in the hidden layer of the network. (4) The association rules generated by ARNN are similar to those generated by traditional association analysis.
Keywords: Artificial neural network;association analysis;association rule;data mining;
瀏覽次數: 16171 下載次數: 233
引用 導入Endnote
頁: 51-78
日期: 2008/10
摘要: 本研究所提出的關聯推理神經網路(Association Reasoning Neural Networks, ARNN)是修改自倒傳遞神經網路演算法，可以產生關聯規則，為傳統的關聯分析開啟完全不同的研究途徑。經由一個數值例題與一個實際例題的結果歸納出以下結論：(1)ARNN的推論輸出值與資料中的信賴度大約相等。(2)當ARNN的隱藏單元數減少時，檢出率、準確率會跟著變小。(3)ARNN可以利用減少隱藏單元數來避免產生信賴度過低的關聯規則。(4)ARNN產生的規則與由傳統關聯分析法所找到的規則重疊性很高，且能找出一些被傳統關聯分析法忽略的關聯規則。
關鍵字: 資料探勘;類神經網路;關聯規則;關聯分析;
Association Reasoning Neural Networks
Abstract: In this study, we proposed the Association Reasoning Neural Network (ARNN) which is derived from ANN, can produce association rules, and open a new approach for association analysis. Based on a numerical and a practical example, some conclusions can be gotten: (1) The reasoning output value of ARNN is proportional to the confidence of the association rules hidden in the data set. (2) As the number of neurons in the hidden layer of ARNN is decreasing, the detection rate and the accuracy rate are decreasing. (3) The ARNN can avoid producing the association rules with low support value by reducing the number of neurons in the hidden layer of the network. (4) The association rules generated by ARNN are similar to those generated by traditional association analysis.
Keywords: Artificial neural network;association analysis;association rule;data mining;
瀏覽次數: 16171 下載次數: 233
引用 導入Endnote