結合決策樹與倒傳遞類神經網路預測台灣太陽能產業之企業績效

Abstract

本研究為協助台灣太陽能公司了解其關鍵績效指標以改善公司績效，使用具有迴歸技術的方法來簡化平衡計分卡(BSC)的變數，並結合決策樹與倒傳遞類神經網路以預測台灣太陽能產業之企業績效。有效解決以下關鍵問題：（1）通過調整關鍵變數(增減5%)了解績效指標改善程度、（2）如何披露相關的關鍵績效指標（KPI）和具體的成果之間的因果關係。儘管已有許多相關研究著重於這類問題，但大多是依賴專家的評估，而本研究使用數據驅動的方法，以了解關鍵變數對績效指標的影響。 由決策樹與倒傳遞類神經網路針對太陽能公司進行績效評估後，發現在分類這部分，所找出的關鍵變數為主力產品的產能利用率與外銷比率。而在預測方面，所找出的關鍵變數為主力產品的銷值比重、產能利用與外銷比率。

This study is to help Taiwanese Solar Industries to understand their key performance indicators to improve organizational performance. Regression was used to simplify variables of the Balanced Scorecard (BSC), and combining Decision Tree with Backpropagation Neural Networks to forecast corporate performances of Taiwanese solar industries. The following critical issues are well addressed: (1) How to improve an outcome to the present degree (i.e. increasing 5%)? (2) How to disclose the underlying causalities between key performance indicators (KPIs) and specific outcomes? Despite numerous studies have been presented to tackle this problem, however, most of them are fully reliant on experts’ assessments. To assist Taiwanese Solar Industries, we focus on the importance weights of KPIs and their causal impacts on outcomes by data-driven methods. Using Decision Tree and Back Propagation Neural Network evaluate performance for solar companies, in this part of the classification, Capacity utilization of major products and Export ratio of major products are key performance indicators. in this part of the prediction, Sale proportion of major products, Capacity utilization of major products and Export ratio of major products are key performance indicators.