Trajectory tracking of changes digital divide prediction factors in the elderly through machine learning

Area of Science:

• Social Sciences
• Computer Science
• Gerontology

Background:

• The intensifying digital divide among the elderly poses significant daily life challenges due to rapid digital transformation.
• Variations in digital technology use create a digital divide even within the elderly population, necessitating urgent management strategies.

Purpose of the Study:

• To predict the digital divide in the elderly population using machine learning techniques.
• To provide insights for managing the digital divide among older adults in an increasingly digital society.

Main Methods:

• Utilized the '2020 Report on Digital Information Divide Survey' data, focusing on ten high-importance independent variables.
• Employed logistic regression, SVM, KNN, decision tree, XGBoost, and a convolutional neural network (CNN) for predictive modeling.
• Compared variable importance from 2019 (pre-COVID-19) and 2020 data.

Main Results:

• Key predictors for the elderly digital divide in 2020 included demographic, internet usage, self-efficacy, and social connectedness.
• A CNN-based model achieved the highest prediction accuracy at 80.4%, outperforming XGBoost (79%) and logistic regression (78.3%).
• Variable importance shifted slightly from 2019 to 2020, with social support being highly important in both years.

Conclusions:

• Strengthening practical digital device connections for the elderly is crucial amidst accelerating digital transformation.
• Applying diverse classification algorithms, including CNNs adapted for social science data, enhances prediction accuracy for the digital divide.
• Proposed management strategies emphasize self-efficacy and social connectedness alongside demographic and internet usage factors.