| 摘要: | 台灣「保險業資產占金融機構資產比率」自2004年的17.36%一路上升到2020年達36.32%是近年來的最高點，可見台灣保險產業的重要性，保險除了是一種金融工具促進社會經濟活動與發展外，且尚有對社會制度發揮穩定的重要功能。然，依據「財團法人金融法制暨犯罪防制中心」2020年舉辦的防制保險犯罪研討會指出2019年的保險犯罪黑數估計損失金額達105億元，同時也指出國外經驗約佔理賠金額10%，這個潛在的保險詐欺、逆選擇風險是各保險公司風險控管的重要課題。
本研究以A公司2018年承保案件為資料來源，並以保單生效後二年內或三年內發生非急性之疾病醫療理賠案件做為目標變項，運用客戶基本資料、投保資料、財務資料、體況資料、招攬人員資料、招攬人員理賠支出率等六大面向納入分析探討，藉由機器學習技術嘗試多種分類器建構核保風險預測模型與實驗，以八種分類器的實驗結果而言，從自變項中可發現以附加醫療險個數、主約險種類別、主附險種代碼及附加附約數是最好的預測變項，在分類器的預測模型效能以梯度提升機具有較穩定及正確率較高的預測能力，其AUC有0.71以上的標準，如以CA分數而言，則以邏輯斯迴歸表現最佳，達0.612以上。
期能藉由本研究結果提供個案公司未來建構以數據為基礎進行風險分類的核保風險評估機制的參考，透過核保風險預測模型提供更有效率對核保風險分類進行核保評估作業，提升核保效能，加速自動化作業並能提升客戶的投保體驗與滿意度，進而提升與公司的黏著度與忠誠度。
;Taiwan′s "Ratio of Assets of Insurance Industry to Total Assets of Financial Institutions" has risen from 17.36% in 2004 to 36.32% in 2020, which is the highest point in recent years. This shows the importance of Taiwan′s insurance industry. Insurance is not only a financial tool to facilitate socio-economic activities, but also plays a vital role in stabilizing social system. According to the seminar on the prevention of insurance crime held by the "Institute of Financial Law and Crime Prevention" in 2020, it is pointed out that the estimated loss of insurance crime in 2019 will reach NT$10.5 billion, and it is also pointed out that foreign experience accounts for about 10% of the claim amount. , this potential insurance fraud and adverse selection risk is an important subject of risk control for insurance companies.
This study uses the underwriting cases of Company A in 2018 as the data source, and takes the non-acute medical claims cases within two or three years after the policy takes effect as the target variable. Analyze and discuss six aspects of customer basic information, insurance information, financial information, physical condition information, agent information, and agent claim settlement rate, and use machine learning technology to try eight different classifiers to build underwriting risk prediction models and experiments. From the experimental results of the seven classifiers, it can be seen from the independent variables that the number of medical insurance, the type of primary insurance, the code of the primary insurance and the number of riders under the policy are the best predictors. The performance of the prediction model is based on the gradient boosting machine, which has relatively stable and high prediction ability, and its AUC reaches more than 0.71. For CA scores, logistic regression is the best, reaching above 0.612.
It is hoped that the results of this study can provide a reference for the case company to construct an underwriting risk assessment mechanism for risk classification based on data in the future, and provide a more efficient underwriting assessment and underwriting risk classification through the underwriting risk prediction model. So as to improve the underwriting risk classification and efficiency, speed up automated operations, and improve customer experience and satisfaction, thereby enhancing customer’s stickiness and loyalty with the company. |
