Targeted Retirement and Investment Plans for Under-Saved Families: A Machine Learning Approach Using Survey of Consumer Finance Data

This research addresses the critical problem of retirement under-saving among American families by developing a comprehensive predictive machine learning system to identify at-risk households and recommend targeted intervention strategies. The retirement savings crisis affects millions of Americans, with nearly 50% of adults aged 60 and above having income below basic needs thresholds. Traditional approaches to this problem have been largely descriptive, identifying the scope of the crisis without providing actionable tools for intervention. This study bridges that gap by combining advanced machine learning techniques with economic analysis to create a practical decision-support system.Using longitudinal data from the Survey of Consumer Finances (SCF) spanning 1989 to 2022, I developed and validated multiple classification models including Support Vector Machines (SVM), Random Forest, and Extreme Gradient Boosting (XGBoost). The dataset encompasses 72 observations across six age groups and 12 survey waves, with over 40 financial variables including income, assets, debts, and derived financial ratios. The XGBoost model emerged as the best performer, achieving 100% test accuracy on optimal configuration and 93.33% mean accuracy across rigorous multi-seed validation with a ROC-AUC of 0.982, demonstrating excellent discriminative ability and model stability.The study makes three major contributions to the literature. First, feature importance analysis revealed that Home-Secured Debt (mortgage debt) is the strongest predictor of retirement preparedness with an importance score of 0.35, followed by Non-financial Assets (0.18) and Debt-to-Income Ratio (0.10). This finding challenges conventional wisdom that income level is the primary determinant of retirement readiness and suggests that housing decisions critically affect long-term financial security. Second, K-Means cluster analysis identified three distinct under-saver segments requiring different intervention strategies: AT RISK families needing automatic enrollment, MODERATE families requiring debt consolidation, and CRITICAL families needing urgent catch-up strategies. Third, scenario modeling quantified the value of early intervention, demonstrating that families under 35 can accumulate approximately $708,000 more in retirement wealth through aggressive saving strategies compared to maintaining current trajectories.The practical implications are significant for financial advisors, employers, and policymakers. This research provides a validated, deployable system for identifying at-risk families, segmenting them into actionable groups, and calculating the return on investment for various intervention strategies. The findings suggest that the retirement savings crisis, while severe, is addressable through targeted, data-driven interventions that account for heterogeneity in family financial situations.