隨著資訊科技進步，資訊獲取的更加便利，因此在這個經濟發展繁榮的環境裡，如何建立一個準確率高的偵測模型，以提供投資者做出正確的投資決策是一項重要的課題。有鑑於過去國內外學者探討盈餘管理相關之議題皆著重於財務方面之數值型態之資訊，因此本研究將加入非財務性之財務報表易讀性指標與未加入財務報表易讀性指標於盈餘管理進行預測，研究結果顯示加入財務報表易讀性指標之模型預測能力較佳。此外本研究使用較新穎之人工智慧技術急速支援向量機(Extreme Support Vector Machine, ESVM) ，此模型將結構風險最小化，在高維度特徵空間中尋找擁有最大邊界之超平面，解決分類問題，有效地區分兩種不同類型之資料。本研究資料取自台灣經濟新報資料庫(Taiwan Economic Journal, TEJ)，研究對象以2017年台灣上市上櫃電子業為研究對象。實證結果顯示將財務報表易讀性投入盈餘管理之預測具有重要性，使投資者做出正確的決策。
According to the progress of information technology and the convenience of information acquisition, it is an important topic to establish a high accuracy detection model in order to provide investors with correct investment decisions in this environment of economic prosperity and prosperity. In view of past domestic and foreign scholars discussion on earnings management related issues, all of them focus on information about financial value types. Therefore, this study will add non-financial statements readability index and financial statements readability index to predict earnings management. In addition, the Extreme Support Vector Machine (ESVM) is used in this study to minimize the structural risk. In the high dimensional feature space, we find the hyper-plane with the largest boundary, solve the classification problem, and effectively divide two different types of data. The research materials are taken from the Taiwan Economic Journal (TEJ) in Taiwan. The subjects of the study are the listed electronics industry in Taiwan in 2017. The empirical results show that it is important to predict the readability of financial statements into earnings management, so that investors can make the right decisions.
