本研究針對國內曾經重編財務報表之公司為研究對象,利用資料探勘技術與多元的角度尋找財務報表重編之公司同特質,以台灣2002~2012年間之上市櫃公司作為樣本,為了達到集群分析之準確性,以三種分群演算法(K平均數法、兩階段法、自我組織映射類神經網路分析)來進行分群之研究,且為驗證模型之準確度,將樣本資料切割,以2002~2006年之資料作為訓練集建立模型,2007~2012年之資料做為測試集之方式驗證,找出一最佳分群模型。最後將全部的樣本投入最佳分群模型,將分群結果所得進行決策樹分析,尋找重編財務報表公司之共同特質。 本研究結果以K平均數法分三群為最佳分群模型,決策樹分析發現重編財務報表公司之特性歸納為三點:(1)會計師發布對該公司繼續經營有疑慮之意見、(2)簽證會計師事務所非四大、(3)股東權益報酬率偏低,顯示若公司具此三項特性之一時,重編財務報表之可能性較高。 關鍵字:資料探勘、分群、決策樹、財務報表重編、公司特質、公司治理、財務指標、審計品質、企業風險管理、盈餘管理
In this study, we focus on domestic companies haing restated financial restatements, using of data mining technology with multi-angle. We intend to find the company’s characteristics of the financial statement restatement. Researching sample includes companies listed in Taiwan market come from 2002 to 2012. In order to improve the accuracy in the cluster analysis, we use three kind of clustering algorithms (K-means, Two-step, SOM). To verify the accuracy of model, the sample data will be divided into the data from 2002 to 2006 as a training set to build model, and from 2007 to 2012 as a testing period to verify the stability of model. After finding the optimal clustering model, all samples will be put into the optimal clustering model. Further, the results of clustering were analyzed by decision tree analysis to find the characteristics of firms with financial statement restatement. We find that, the optimal clustering model is K-Means method divided into three groups. In the three group, decision tree analysis found that firms with financial restatement has the follow characteristics: (1) accountants issued that ability of company continuing operating is doubtful, (2) non-big four CPA firm, (3) lower return on equity. When the company has displayed one of these three characteristics, the higher the likelihood of restatement of financial statements Keywords: Data Mining, Clustering, Decision trees, Restated Financial Statements, Company Characteristics, Corporate Governance, Financial Indicators, Audit Quality, Enterprise Risk Management, Earnings Management