探討機器學習於採購舞弊偵測設計：以A公司為例

Abstract

A公司透過企業併購擴張事業版圖，成為兩岸三地的巨型企業之一。隨著業務複雜度增加及公司組織擴大，內部控制與監督管理未適時隨著組織及業務擴張而及時整合，衍生風險管理議題。本論文考量下列各項因素以探討A公司透過機器學習偵測採購舞弊的設計： 1. A公司管理組織龐大，採購交易頻率高，交易金額與數量龐大，適合採行先進科技輔佐採購風險管理。 2. A公司內部採行不同ERP系統平台，並且儲存大量交易資料數據，適合應用資料分析與機器學習技術偵測採購舞弊並管理相關風險。 3. A公司中的採購舞弊行為模式係因為累積多年的內部控制疏漏演變而成，且不易偵測，但是與國際上相關文獻研究範疇類似，適合採用機器學習進行採購風險偵測應用之設計與研究探討。 4. A公司所屬產業主要為生產製造領域，採購作業為核心經營流程之一，與多數台灣企業屬性類似。根據過去顧問服務經驗，A公司中發生的舞弊態樣亦常見發生於台灣企業中，因此研究探討A公司採購舞弊偵測模式，亦可以提供做為其它台灣企業的借鏡參考。 本論文參考國內外機器學習相關文獻，及參考過去在A公司提供服務的實際經驗，說明A公司內實際發生之舞弊手法，以資料分析方法說明個案公司存在之特定採購舞弊活動偵測應用方法，並提出若採用機器學習方法的架構設計。

A Company has expanded its business territory through corporate mergers and acquisitions in the past, becoming one of the giant enterprises in the Greater China region. With the increasing complexity of operations and the expansion of the company's organization, internal control and supervisory management have not been timely integrated as the organizational business expands, giving rise to risk management issues. This thesis considers the following factors to explore A Company's design for detecting procurement fraud through machine learning: 1. A Company has a large management organization, high frequency of procurement transactions, and large transaction amounts and quantities, making it suitable for adopting advanced technology to assist in procurement risk management. 2. A Company adopts different ERP system platforms internally and stores a large amount of data, making it suitable for applying data analysis and machine learning techniques to detect procurement fraud and manage related risks. 3. The pattern of procurement fraud behavior in A Company has evolved from years of internal control loopholes and is difficult to detect, but it is similar to the scope of international literature research, making it suitable for research and exploration of machine learning application designs. 4. A Company's main industry is in the manufacturing sector, with procurement operations being one of the core processes, similar to most Taiwanese enterprises. Based on past consulting service experience, the patterns of fraud occurring in A Company are also commonly seen in Taiwanese enterprises. Therefore, studying A Company's procurement fraud detection model can also provide reference for other Taiwanese enterprises. This thesis references domestic and foreign literature on machine learning, as well as past practical experiences in providing services to the case company, to illustrate the actual fraudulent practices within the case company. It explains specific procurement fraud detection application methods using data analysis methods and proposes a framework design if machine learning methods are adopted.