Comparing count regression and machine learning in automobile insurance claims prediction
Author Information
Author(s): Gadir Alomair
Primary Institution: Department of Quantitative Methods, School of Business, King Faisal University, Al-Ahsa, Saudi Arabia
Hypothesis
This study investigates the utility of machine learning in improving forecast accuracy under conditions of zero-inflation in automobile insurance claims data.
Conclusion
The SVM model outperforms others in predictive accuracy, particularly in handling zero-inflation, followed by the ZIP and ZINB models.
Supporting Evidence
- The SVM model demonstrated superior performance compared to RF and ANN models, with a testing MAE of 0.854996.
- The Poisson and NB models exhibited the weakest performance among the regression models.
- Both count regression and ML models showed that SVM outperformed all others.
- The ZIP and ZINB models demonstrated a slight edge in performance compared to other count regression models.
Takeaway
This study looks at how well different models can predict car insurance claims, especially when many people don't make claims. It found that a machine learning model called SVM is the best at making these predictions.
Methodology
The study involved a comparative evaluation of several models, including Poisson, NB, ZIP, hurdle Poisson, ZINB, random forest, support vector machine, and artificial neural network on an insurance dataset, assessed using mean absolute error.
Limitations
The study is based on a single dataset, which may limit the generalizability of the findings.
Participant Demographics
The dataset includes demographic and socioeconomic information about policyholders, such as gender, age, marital status, and income.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website