以機器學習為基礎之車禍致死案件精神損害賠償判決預測

Legal Judgment Prediction of Solatium for Fatal Car Accident Cases Based on Machine Learning

謝德成(De-Cheng Hsieh)

指導教授 ： 陳履恒

國立暨南國際大學/科技學院/資訊工程學系/碩士(2021年)

https://doi.org/10.6837/ncnu202100116

摘要

近年來，基於機器學習的司法判決預測已然成為了法學人工智慧領域中的重要研究議題，其原因在於它可以作為一個有用且強固的輔助工具，來幫助律師與法官們更快、更準確地預測法律案件的可能判決結果，同時讓他們的工作更有效率。
而本研究試圖預測臺灣車禍致死案件精神損害賠償的判決金額，並藉由K最近鄰、CART決策樹以及隨機森林等機器學習演算法，來建立預測模型。此外，本研究也期望藉由不純度重要性的計算，來找出個別審酌因素特徵對於判決慰撫金額的重要性與影響力。同時，也嘗試透過相關係數來檢視個別審酌因素特徵與判決慰撫金額之間的相關性。再者，本研究也根據個別最佳模型所計算出的測試誤差離群值，來偵測哪些判決的慰撫金額已相當程度地脫離多數主流判決，而顯然有偏向原告或被告之虞，因此屬於非多數主流之離群判決。
研究結果顯示，本研究所建立之最終最佳模型具有強固優良的預測效能，且本研究更進一步揭示了個別審酌因素特徵對於判決慰撫金額的特徵重要性與影響力排名以及個別審酌因素特徵與判決慰撫金額之間的相關性。此外，本研究也具體揭露出非多數主流之離群判決。

關鍵字

司法判決預測 ； 機器學習 ； K最近鄰 ； CART決策樹 ； 隨機森林 ； 不純度重要性 ； 相關係數 ； 測試誤差

並列摘要

Recently, legal judgment prediction using machine learning techniques has been a very important issue of artificial intelligence in legal domain since it can be a useful and robust tool to help lawyers and judges predict the potential judgment of a legal case rapidly and precisely and make their work more efficient.
In this paper, we aim at predicting the legal judgment of solatium for fatal car accident cases in Taiwan. Therefore, we use K-Nearest Neighbor, Classification and Regression Tree, and Random Forest as regressors to build the optimal predictive model.
Furthermore, we endeavor to find out the feature importance among the legal factors by rank and the correlation between the legal factors and the discretionary damages of judgments with the help of the impurity importance and the correlation coefficient. In addition, we try to reveal the judgments which are contrary to the mainstream with the outliers of the testing error.
The experimental results show that our optimal model performs well and achieves a strong performance. Besides, we successfully present the feature importance among the legal factors, the correlation between the legal factors and the discretionary damages of judgments, and the judgments which are contrary to the mainstream.

並列關鍵字

Legal Judgment Prediction ； Machine Learning ； K-Nearest Neighbor ； Classification and Regression Tree ； Random Forest ； Impurity Importance ； Correlation Coefficient ； Testing Error

參考文獻

[1] Y. Chang, H. Ho and T. Li, "An Empirical Study of Pain and Suffering Damages in Fatal Car Accident Cases in Taiwan," Chengchi Law Review, vol. 149, pp. 139-219, 2017.

[2] H. Surden, "Machine Learning and Law," Washington Law Review, vol. 89, no. 1, pp. 87-115, 2014.

[3] D. M. Katz, "Quantitative legal prediction-or-how I learned to stop worrying and start preparing for the data-driven future of the legal services industry," Emory Law Journal, vol. 62, no. 4, pp. 909-966, 2013.

[4] H. Zhong, Z. Guo, C. Tu, C. Xiao, Z. Liu and M. Sun, "Legal Judgment Prediction via Topological Learning," in Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, 2018.

[5] T. W. Ruger, P. T. Kim, A. D. Martin and K. M. Quinn, "The SupremeCourt Forecasting Project: Legal and Political Science Approaches to Predicting Supreme Court Decisionmaking," Columbia Law Review, vol. 104, no. 4, pp. 1150-1210, 2004.