應用機器學習演算法預測首次代幣發行詐欺之可能;Applying machine learning algorithms to predict the probability of ICO fraud.

NCUIR > School of Management at National Central University > Research institute of accounting > Electronic Thesis & Dissertation > Item 987654321/95692

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/95692

Title:	應用機器學習演算法預測首次代幣發行詐欺之可能;Applying machine learning algorithms to predict the probability of ICO fraud.
Authors:	吳秉純;Wu, Bing-Chuen
Contributors:	會計研究所
Keywords:	首次代幣發行;詐欺預測;機器學習;initial coin offering;fraud prediction;machine learning
Date:	2024-07-22
Issue Date:	2024-10-09 17:09:40 (UTC+8)
Publisher:	國立中央大學
Abstract:	摘要首次代幣發行（ICO）是一種由區塊鏈和加密貨幣所衍生出來的募資方式，相較於首次公開募股（IPO），ICO利用區塊鏈的特性提升了交易的透明度及速度，不僅節省了複雜的申請流程，還有機會能吸引到全球投資者的資金。但隨著ICO發行數量的增加，利用其行使詐騙的案例層出不窮。因為ICO沒有強制揭露資訊的規定，發行方擁有決定資訊揭露程度的權利，因此在事前投資人大多只能從ICO的網站及白皮書取得ICO相關資訊。為了解決ICO投資者頻繁遭受詐騙的情況，本研究利用白皮書資訊建立預測模型來預測ICO為詐欺的可能性。本研究採用Random Forest、KNN、SVM、Naïve Bayes、Probit作為預測模型，以2016-2020年ICO盛行期間的白皮書作為研究樣本，並進一步了解白皮書內提供的各項資訊對ICO詐欺的影響程度。研究結果顯示，五種預測模型中SVM的預測表現最佳，且在本研究用來預測ICO詐騙的特徵中以詳述智慧合約與願景藍圖兩項資訊對案例的影響最顯著，代表這兩項資訊最能事前預測案例是否未來可能產生舞弊。本研究提供未來ICO的投資者一種新的評估方式，使其降低遭受詐騙的可能性。關鍵詞:首次代幣發行、詐欺預測、機器學習 ;Abstract Initial Coin Offering (ICO) is a fundraising method derived from blockchain and cryptocurrency. Compared to Initial Public Offering (IPO) , ICO make good use of blockchain to enhance transaction transparency and speed, which not only avoid the complex application process but also has the potential to attract funds from global investors. However, with the increasing number of ICOs, cases of fraud have proliferated. Since ICOs are not required to disclose information compulsorily, issuers have the right to determine the extent of information disclosure. Therefore, investors often can only obtain related information from the ICO’s website and white paper before investing. To address the frequent fraud faced by ICO investors, this study establishes a prediction model using white paper information to predict the likelihood of an ICO being fraudulent. The study employs Random Forest, KNN, SVM, Naïve Bayes, and Probit as prediction models, using white papers of ICOs between 2016 and 2020 as research samples, and further investigates the impact of various information provided in the white papers on ICO fraud. The results show that among the five prediction models, SVM performs the best. Additionally, among the features used to predict ICO fraud in this study, detailed information on smart contracts and vision blueprints had the most significant impact on the result, indicating that these two pieces of information are the most indicative when it comes to predicting future fraud. This study provides a new evaluation method for future ICO investors, reducing the likelihood of investors getting defrauded. Keywords: initial coin offering, fraud prediction, machine learning
Appears in Collections:	[Research institute of accounting ] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	24	View/Open

社群 sharing

Loading...