以機器學習方法解決保險理賠數據集不平衡之問題

Abstract

近年來，金融科技提供了新形態的金融服務，給傳統的金融產業帶來了全面性的衝擊。保險業作為金融業的一部分，也因應了金融科技創新的潮流，提出了保險科技，包含了網路平台式的經營模式，與透過機器學習來分析數據。本篇研究將針對保險業中常見的資料集，車輛（包含機車與汽車）碰撞情況報告，使用機器學習方法進行分析，並預測車禍發生後傷害的嚴重程度。然而，在資料集當中比較少的死亡車禍案例，才是我們真正需要去在意的。在保險理賠中，死亡車禍會讓保險公司付出大量的金錢。因此，我們需要去進一步提升對於死亡車禍的預測。為了衡量我們預測的結果，本文中採用了Precision和Recall來取代Accuracy，著重在死亡車禍的判斷上。最後，我們會探討本研究在保險服務中的應用。

In recent years, financial technology has provided a novel form of financial services, which has brought a comprehensive impact to the traditional financial market. The insurance sector, which is an important part of the financial sector, has developed InsurTech, including digital finance and machine learning. This paper will focus on the data sets commonly used in the insurance sector, such as collision reports of vehicles (including motorcycles and cars). We use machine learning methods to analyze and predict the severity of injuries after a car accident. However, the few cases of fatal crash in the data set are what we really need to care about. In insurance claims, a fatal crash will cost the insurance company lots of money. Therefore, we need to further improve the prediction of the fatal crash. In order to measure the results of our predictions, Precision and Recall are used in this paper to replace Accuracy, focusing on the judgment of fatal crash. Finally, we will explore further application of this research in insurance services.