Fear and Finance: An Unsupervised Machine Learning Study on Credit-Averse Households in the U.S

Household financial health is a crucial topic for policymakers and credit companies. This study used data from the Survey of Consumer Finances (SCF) to cluster credit-fearful households based on their financial profiles using feature selection and principal component analysis (PCA) techniques. The researchers first used the trimmed variance method to select five high-variance features: home value, assets, debt, net worth, and age. Then, they employed the K-means algorithm to cluster the households into four distinct groups based partly on net worth. The researchers fitted the final model using a pipeline consisting of StandardScaler and K-Means. They visualized the results using Plotly Express, creating a scatter plot to visualize the distribution of households across different groups. The findings revealed that the difference between the clusters primarily lies in home values. The households with the highest net worth do not necessarily carry the highest debt. Additionally, households with high net worth and assets may not have high home values, while households with high home values may have relatively lower assets. The study provides actionable insights that can guide credit companies to adopt an algorithmic approach to understand better and serve their customers. The results have practical implications for credit unions that seek to identify credit-fearful households in the SCF dataset and improve their financial well-being.