Predictive Analysis of Automobile Insurance on Time Series Data

• Main Authors: Hui-Yu Yu, 游輝育
• Other Authors: K. Robert Lai
• Format: Others
• Language: zh-TW
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/33994781090283444383

碩士 === 元智大學 === 資訊工程學系 === 105 === Automobile insurers survey that call for the use of usage-based-insurance has been an emerging service as an automobile insurance to set the premium individually to each policyholder. Personalized automobile insurance mechanism posts challenges that are different from general driving behaviors analyzing in driver assistance applications. In this paper, a novel framework based on the boosted multiple kernel learning is proposed to reflect a driving risk level to each individual driver for automobile usage-based-insurance. In the proposed framework, a set of kernels is specified to represent the driving-oriented, driver-oriented, and lane-oriented attributes. These multiple kernels are carefully integrated by the AdaBoost technique to realize predicting. Experimental results on lab-recorded driving dataset in real-world conditions show the reliable performance of the proposed framework in terms of using different number of kernel scales and usefulness of different attributes.