Studies of life satisfaction in older adults have been conducted extensively through empirical research, questionnaires, and theoretical analysis, with the majority of these studies basing their analyses on simple linear relationships between variables. However, most real-life relationships are complex and cannot be approximated with simple correlations. Here, we first investigate predictors correlated with life satisfaction in older adults. Then, machine learning is used to generate several predictive models based on a large sample of older adults (age ≥ 50 years; n = 34,630) from the RAND Health and Retirement Study. Results show that subjective social status, positive emotions, and negative emotions are the most critical predictors of life satisfaction. The Support Vector Regression (SVR) model exhibited the highest prediction accuracy for life satisfaction in older individuals among several models, including Multiple Linear Regression (MLR), Ridge Regression (RR), Least Absolute Shrinkage and Selection Operator Regression (LASSO), K Nearest Neighbors (KNN), and Decision Tree Regression (DT) models. Although the KNN and DT models exhibited better model fitting than MLR, RR, and LASSO, their performances were poor in terms of model validation and model generalization. These results indicate that machine learning is superior to simple correlations for understanding life satisfaction among older adults.
Keywords: life satisfaction; machine learning; older adults; predictive models.