- 期刊
視覺化與機器學習協同整合之增強分析－以風險視覺化為例
The Augmented Analytics Integrated with Visualization and Machine Learning: Illustrated by Risk Visualization
摘要
近年來人工智慧蔚為風潮，機器學習可以加速資料分析或提高決策結果準確度，但如何才能有效提升人類決策能力呢？機器學習是否可能成為決策黑箱，造成決策的盲點，或是導致使用者的抗拒？為了彌補機器學習在解釋上的不足，本研究將機器學習所挖掘之繼續經營疑慮法則，進一步以學者提出的視覺化分析知識生成模型及結合杜邦分析法之財務指標，並以風險視覺化儀表板呈現分析結果；後續則透過使用者訪談，評估與回饋視覺化介面的決策效益。研究目的在於評估系統使用者在分析財務報表過程中，藉由操作互動式視覺化圖表是否有利於理解及驗證機器學習模型所辨識之財務危機決策規則。目的之二則是探討使用者如何結合視覺化工具與機器學習模型，達到兼俱正確性與可解讀性之綜合效果。本研究經過使用者評估發現，應用視覺化與機器學習協同整合之視覺化分析知識生成模型，所產製的財務預警儀表板，確實有助於使用者增強理解機器學習分析結果，並可深入分析各指標變動趨勢，促使提升分析人員對資料之敏感度，提供證據以即時採取相應之措施。
並列摘要
Artificial intelligence has become a trend in recent years. Machine learning can accelerate data analysis or improve the accuracy of decision-making results. But how can it effectively improve human decision-making capabilities? Is it possible that machine learning can become a decision-making black box, causing blind spots in decision-making, or causing users' resist? In order to make up for the deficiencies of machine learning in explanation, this research applies the rule of continuous operation doubts excavated by machine learning process, and adopts the theory of Knowledge Generation Model for Visual Analytics combined with the financial indicators of DuPont analysis method, and present them with financial risk dashboard; follow-up through user interviews to evaluate and feedback the decision-making benefits of the visual interface. The purpose of the research is to evaluate whether system users can understand and verify the financial crisis decision-making rules identified by the machine learning model by operating interactive visual dashboards in the process of business analytics. The second purpose is to explore how users can combine visualization tools and machine learning models to achieve a comprehensive effect of correctness and interpretability. Through user evaluation, this research found that the application of the Knowledge Generation Model for Visual Analytics that integrates visualization and machine learning to produce financial risk dashboards does help users to enhance their understanding of the results of machine learning analysis and can provide in-depth analysis. The trend of changes in various indicators urges analysts to increase their sensitivity to data and provide evidence to take immediate measures.