Leveraging machine learning to decode insurance purchasing disparities in Canadian households: A PCA approach

This thesis investigates the factors influencing insurance spending among Canadian households, employing advanced machine learning techniques and Principal Component Analysis (PCA). This research develops an integrated predictive model to forecast household expenditures on life, health, and auto insurance, incorporating a comprehensive range of determinants such as household characteristics, economic conditions, and regional differences. Utilizing a robust dataset from the Survey of Household Spending (SHS) for the years 2010 to 2017, with 2019 serving as the validation year, the study applies PCA to manage high-dimensional data effectively, thereby enhancing the predictive performance of the machine learning algorithms used. The results indicate that the model predicts insurance expenditures with notable accuracy; however, it slightly underestimates life insurance costs with an actual expenditure of $1,381 compared to the predicted $1,263 while providing highly accurate forecasts for health insurance. The predictions for car insurance expenditures exhibit larger variances. The findings highlight the substantial benefits of integrating PCA and machine learning to advance predictive analytics in the insurance industry. The study offers critical insights for insurance providers, policymakers, and consumers, laying a data-driven groundwork for strategic decision-making and policy development. Recommendations for future research include refining the predictive models and investigating additional variables that may influence insurance spending. This thesis not only contributes to the academic discourse but also provides actionable strategies to enhance the accuracy and efficacy of forecasting models in the insurance sector.