基於集成學習與SHAP解釋投資型保單消費者優勢綜合評估

Abstract

中文摘要 本研究運用數據分析與機器學習技術，評估投資型保單對消費者所提供優勢[1]。研究採用保發中心2023年至2025年第一季的投資型保險統計數據，涵蓋保費收入、保險給付、帳戶價值與新契約四大面向，針對變額壽險、變額萬能壽險及變額年金三大產品進行分析[2]。研究過程中先建立高維特徵矩陣，透過特徵工程創建多項績效評估指標，包括保費對理賠比、年化帳戶價值增長率、投資效率指標等關鍵衡量面向[3]。接著運用嶺迴歸(L2迴歸)、隨機森林、深度神經網路等多種機器學習模型進行集成學習，透過Weighted Voting及Stacking訓練並預測分析，再透過K-means非監督式聚類方法對消費者群體進行市場細分，並使用SHAP值分析關鍵影響因素[4]。 研究結果顯示，投資型保單相較傳統保單具顯著優勢：投資彈性優勢提供多投資標選擇權，費用透明平均總費用率2.5%-3.5%且明確揭露，專設帳簿提供破產隔離保護[5][6]。三大產品以變額年金表現最佳，保費對理賠比達1.33，年化帳戶價值增長率9.3%，長期持有效率14.8%[7]。消費者分為高淨值群體（15%，貢獻40%保費）、年輕族群（48%）、穩健型群體（37%）三類，其中高淨值群體年化效率可達23.5%1。本研究對於不同風險偏好消費者建立決策矩陣，可供保險業者提供產品優化建議，以提升台灣投資型保險之市場發展[8][9]。

Abstract This study employs data analytics and machine learning techniques to evaluate the advantages that investment-linked insurance policies provide to consumers1. The research utilizes investment-linked insurance statistical data from the Taiwan Insurance Institute from 2023 to the first quarter of 2025, encompassing four key dimensions: premium income, insurance benefits, account value, and new contracts, with analysis focused on three major products: variable life insurance, variable universal life insurance, and variable annuities2. The research process initially established a high-dimensional feature matrix, creating multiple performance evaluation indicators through feature engineering, including premium-to-benefit ratios, annualized account value growth rates, and investment efficiency indicators, among other key measurement dimensions3. Subsequently, ensemble learning was implemented using various machine learning models including ridge regression (L2 regression), random forest, and deep neural networks through weighted voting and stacking for training and predictive analysis, followed by unsupervised K-means clustering methods for consumer market segmentation, with SHAP values employed to analyze key influencing factors4. The research findings demonstrate that investment-linked insurance policies possess significant advantages over traditional insurance policies: investment flexibility advantages provide multiple investment target selection rights, fee transparency with average total expense ratios of 2.5%-3.5% with clear disclosure, and separate accounts providing bankruptcy isolation protection5 6.Among the three major products, variable annuities demonstrated superior performance with premium-to-benefit ratios reaching 1.33, annualized account value growth rates of 9.3%, and long-term holding efficiency of 14.8%1. Consumers were segmented into three categories: high net worth individuals (15%, contributing 40% of premiums), young demographics (48%), and conservative investors (37%), with high net worth individuals achieving annualized efficiency rates of up to 23.5%2. This study established decision matrices for consumers with different risk preferences, providing product optimization recommendations for insurance companies to enhance Taiwan's investment-linked insurance market development7 8.