A prediction method for older adult care service demand combining improved RF algorithm and logistic regression

TL;DR

This paper proposes a new model combining improved random forest and logistic regression to accurately predict care service demand for older adults.

Contribution

The novel integration of an adaptive feature selection in RF with LR improves prediction accuracy and interpretability for older adult care services.

Findings

The model achieved 95.30% accuracy in predicting care service demand.

It outperformed standalone RF and LR models in recall, F1 score, and AUC.

The method provides reliable decision support for resource allocation in elderly care.

Abstract

With the acceleration of global aging, the accurate prediction of care service demand for older adults is of significant importance for optimizing resource allocation. Traditional prediction methods often lack sufficient accuracy when dealing with high-dimensional and nonlinear health data.

A prediction model integrating an improved random forest (RF) algorithm and logistic regression (LR) is proposed. The method introduces an adaptive feature selection strategy within the RF framework to dynamically select the most influential feature subsets. Key features screened by the optimized RF are then used to construct an LR classifier, leveraging its strengths in handling linear relationships and providing interpretability.

The proposed model achieved an accuracy of 95.30%, a recall rate of 92.60%, an F1 score of 93.90%, and an area under the receiver operating characteristic curve of 0.934 in predicting care service demand for older adults. These results were significantly better than those obtained using the RF or LR models alone.

The findings indicate that the integrated approach effectively improves prediction accuracy and reliability. The model offers a robust decision-support tool for care service institutions and government departments in resource planning and service allocation for the older population.

Figures