查詢結果分析
來源資料
相關文獻
- 可解釋人工智慧在財務風險預警之應用:以臺灣AI概念股為例
- 基於集成學習框架之信用違約預測--以信用卡客戶為例
- 結合機器學習與時間序列模型預測短期負載
- 集成學習於臺灣邊境食品安全之風險預測
- 複合特徵工程結合機器學習於信用風險評估決策支持系統之研究
- Genetic Programming for Classification of Remote Sensing Data
- 空間配置規則的推導--人工智慧中案例學習法之運用
- Comparison of ID3 Symbolic Learning and Neural Network for Selecting Deep Foundation Type
- 地理資訊擷取之最佳化方法--以道路養護為例
- 資料採礦中決策樹的理論及應用
頁籤選單縮合
| 題 名 | 可解釋人工智慧在財務風險預警之應用:以臺灣AI概念股為例=Application of Explainable Artificial Intelligence in Financial Risk Early Warning: Evidence from Taiwan's AI Concept Stocks |
|---|---|
| 作 者 | 詹家榜; 陳逸凡; | 書刊名 | 正修學報 |
| 卷 期 | 38 2025.12[民114.12] |
| 頁 次 | 頁434-453 |
| 分類號 | 312.83 |
| 關鍵詞 | 機器學習; 財務困境預警; 人工智慧概念股; SHAP分析; 集成學習; Machine learning; Financial distress prediction; AI-related stocks; SHAP analysis; Ensemble learning; |
| 語 文 | 中文(Chinese) |
| 中文摘要 | 隨著人工智慧技術的快速發展,台灣 AI 相關企業在全球供應鏈中扮演關鍵角色,然而快速 的股價上漲亦引發潛在財務風險之關注。本研究建構基於機器學習技術的財務困境預警模型, 以台灣 AI 概念股為研究對象,涵蓋 2018 年至 2024 年期間之財務數據。研究採用台灣經濟新報 資料庫之上市公司財務資訊,透過篩選關鍵財務指標作為預測變數,比較十種機器學習演算法 的預測效能。 實證結果顯示,機器學習技術在財務困境預測上展現顯著效果,整體準確率普遍達到 85% 以上,AUC 值超過 0.93。其中梯度提升機演算法表現最為優異,在準確率(0.950)、精確率 (0.951)、召回率(0.950)、F1 分數(0.950)及 AUC 值(0.991)等評估指標上均呈現卓越表現。透過 SHAP 值分析技術,本研究深入探討各項財務指標對預測結果的影響程度,實現模型決策過程 的透明化。研究發現集成學習方法的預測效能明顯優於單一分類器,為財務風險管理提供重要 的決策支援工具。 |
| 英文摘要 | With the rapid advancement of artificial intelligence (AI) technology, Taiwan’s AI-related enterprises play a pivotal role in the global supply chain. However, the concurrent surge in stock valuations has precipitated concerns regarding potential underlying financial risks. This study constructs a financial distress prediction model based on machine learning techniques, focusing on Taiwan’s AI-related listed companies as the research subjects. The analysis encompasses financial data spanning the period from 2018 to 2024. Utilizing financial information from the Taiwan Economic Journal (TEJ) database, this study screens key financial ratios as predictive variables to benchmark the forecasting performance of ten distinct machine learning algorithms. Empirical results demonstrate that machine learning techniques exhibit significant efficacy in predicting financial distress, with overall accuracy generally exceeding 85% and Area Under the Curve (AUC) values surpassing 0.93. Notably, the Gradient Boosting Machine (GBM) algorithm achieved superior performance, demonstrating excellence across all evaluation metrics, including Accuracy (0.950), Precision (0.951), Recall (0.950), F1-score (0.950), and AUC (0.991). By employing SHAP (SHapley Additive exPlanations) value analysis, this study rigorously investigates the impact of individual financial indicators on prediction outcomes, thereby enhancing the transparency of the model's decision-making process. The findings indicate that ensemble learning methods significantly outperform single classifiers in predictive efficacy, providing a critical decision support tool for financial risk management. |
本系統中英文摘要資訊取自各篇刊載內容。