Abstract: Objective To explore the influencing factors and provide reference for the prevention and intervention of cognitive decline in elderly people with subjective cognitive decline (SCD) by comparing cognitive function predictive models based on machine learning. Methods With the data of China Health and Retirement Longitudinal Study (CHARLS) in 2018, 2,969 elderly people with SCD were screened out. The least absolute shrinkage and selection operator (LASSO) regression, support vector regression (SRV), and random forest (RF) regression were used to construct predictive models for cognitive function of the elderly with SCD. The influencing factors of cognitive function were extracted based on the optimal predictive model. Results Among the three models, the one constructed by RF regression demonstrated the highest accuracy in predicting cognitive function in older adults experiencing SCD (R²=0.864, MAE=1.988, MSE=5.879). The factors affecting the cognitive function of the elderly with SCD can be ranked in order of importance as follows: physical dysfunction, age, the total score of depression, self-assessment of health, education background, the total score of entertainment, cleanliness, nap time, total score of IADL, with broadband or not. Conclusion Cognitive function predictive model constructed using RF regression demonstrates superior performance compared to models constructed using LASSO regression and SVR. By integrating the specific circumstances and risk factors of older adults, clinical professionals can develop personalized and multidimensional intervention plans that address factors such as learning, leisure activities, daily routines, nap time, and internet usage, so as to prevent cognitive function decline in older adults experiencing SCD.

Key words: subjective cognitive decline, older adult, machine learning, cognitive function