University of Twente Student Theses


A machine learning approach for modeling frequency and severity

Tuininga, Frits (2022) A machine learning approach for modeling frequency and severity.

[img] PDF
Abstract:Gradient Boosting Regressor (GBR), eXtreme Gradient Booster (XGB), Random Forest (RF) and Neural Network (NN) with specified parameters do not improve or outperform Generalised Linear Model (GLM) when following the frequency-severity method on vehicle insurance data. The four machine learning models are selected due to their explainable results and outstanding performances in related areas (see Section 2). Feature importance and partial dependence plots are made for GBR, XGB and RF to gain more insight into prediction explainability. Furthermore, permutation importance and partial dependence plots are created for NN to acquire a better understanding of prediction explainability. In a nutshell, this research consists of two experiments. The first experiment is divided into four phases: pre-processing, training & testing, importance plot creation and evaluation of risk premium predictions. The second research experiment is concerned with the generalisability of the pre-processing. The generalisability of these phases is demonstrated by running the program on another data set (California Housing Data [1]). By generalising these phases, the same machine learning models can be applied to a range of other data sets within the working environment of company X. In conclusion, our study found that when trained on vehicle insurance data, GBR, XGB, RF, and NN cannot outperform GLM. Nonetheless, when trained on different data sets this approach has the potential of improving or replacing other models. Training the models on new data is relatively easy due to the generalisability of the pre-processing and training & testing phase. Therefore, it is strongly recommended to apply the program on different data sets.
Item Type:Essay (Master)
Achmea, Apeldoorn, Netherlands
Faculty:BMS: Behavioural, Management and Social Sciences
Subject:02 science and culture in general
Programme:Industrial Engineering and Management MSc (60029)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page