本計畫提出一個新穎的公司績效預測模型，此模型採用了網路基礎資料包絡分析法以差 額變數評估、降維技術、整體學習概念與企業生命週期理論。網路基礎資料包絡分析法 可以同時處理多投入與多產出變數，可使評估之準則更具客觀性，亦可克服傳統資料包 絡分析法無法處理決策單位內部結構的問題，因此，採用此架構評估企業的經營績效會 較適切。然而，網路型態資料包絡分析法會遭遇到需處理一重要課題，即過多投入與產 出變數將會損害模型區別能力與判斷效果，本計畫將採用流行學習技術來降低投入與產 出變數之維度，此法可處理非線性之資料結構，故適合用來分析財務變數，因為絕大多 數之財務資料之結構皆為非線性。智慧資本以被廣泛認定是企業未來可否永續發展之關 鍵指標，然而對智慧資本之定義卻沒有統一且一致之標準，故本計畫採用整體學習理論 歸納出具解釋能力之關鍵智慧資本指標，在預測模型採用支援向量機，此模型具有內部 參數少、小樣本具較佳表現等特性，故將其應用於建構企業經營績效預測模型。最後， 將產品生命週期的概念導入企業的經營績效評估當中，進一步分析企業在不同的生命週 期下，智慧資本對其經營績效的影響性，以利管理者以及決策者做有效的規劃與評估。 The study proposed an emerging performance forecasting mechanism which incorporates Network-based data envelopment analysis (NDEA) model with slack-based measure (SBM), manifold learning, intellectual capital and corporate life cycle theory. The NDEA deal with assessments of relative efficiency of decision making units (DMUs) considering multiple inputs and multiple outputs and overcome the drawbacks of traditional DEA to handle the internal structure of the DMUs. However, too many input and output variables will deteriorate the discriminant ability of NDEA. Thus, dimensionality reduction is an inevitable pre-process. ISOMAP, manifold learning, is a dimensionality reduction technique used to determine the useful information embedded into large dataset with non-linear structure. The raw data undergone the dimensionality reduction process fed into NDEA will enhance the discriminant ability and decrease the computational complexity. Intellectual capital is viewed as the most essential source of performance which will sound the competitiveness of corporation. However, the component of intellectual capital does not have the specific definition. Thus, the study based on ensemble learning to determine the most essential component of intellectual capital by numerous combination strategies. The support vector machine (SVM) was used as a base classifier to construct the performance forecasting model. Finally, the study based on corporate life cycle theory to examine the effectiveness of intellectual capital on corporation with dissimilar life cycle stage.
