Leveraging Artificial Intelligence Algorithms for Retirement Income Optimization in Financial Planning

Abstract

Responsibility for one's own financial future is required in people's financialized everyday lives.  However, as a larger proportion of the population ages, more people are concerned about their financial stability due to factors such as fluctuating retirement and pension systems, uncertain government funding, and market and other crises. The present study addresses the urgent issue of low uptake of pension schemes by developing and testing high-performance predictive models based on data from FinAccess. To address the extreme imbalance among the classes, the more extensive AdaSyn resampling method was used. A sophisticated combination of 22 sociodemographic predictors was specified to train and optimize Multi-Layer Perceptron (MLP) and Decision Tree (DT) models systematically. The obtained results show outstanding predictive power, as the MLP model has an accuracy of 95.74%, precision of 96.26%, recall of 95.60%, F1-score of 95.76%, and a near-perfect AUC of 0.9968, while the DT model yielded closely competitive metrics of 95.25% accuracy, 95.65% precision, 95.03% recall, 95.26% F1-score, and a 0.9950 AUC. A comparative study proves that both of the proposed models are far much better than modern standards, such as RNN-LSTM, LightGBM, and Random Forest. The study develops a new analytical pipeline that offers practical, quantitative information of creating specific financial inclusion initiatives to enhance security in changing retirement environments.