Work

A Lightweight Gradient Boosting Model for Predicting Postoperative Fall Risk in the Elderly

Year: 2026
Type: conference-paper
Abstract: Older adults after fracture surgery are at particularly high risk of postoperative falls because of pain, impaired mobility, reduced muscle strength, balance dysfunction, and delayed functional recovery. Falls during the early rehabilitation period may further worsen mobility, prolong hospital stay, increase care dependence, and compromise overall recovery. To address the need for early and efficient risk stratification in this clinically vulnerable population, this study proposes PF-LGBoost, a lightweight gradient boosting model for postoperative fall risk prediction in older adults after fracture surgery. Built upon LightGBM, the model incorporates three modules: adaptive fracture-related feature pruning, a lightweight weak learner, and postoperative sample weighting. Through dynamic feature selection, simplified tree construction, and risk-sensitive weighting, the proposed model balances predictive accuracy with deployment efficiency. Experiments were conducted on 800 samples collected from a public dataset and a multicenter clinical dataset of older adults after fracture surgery. Comparative and ablation studies with 5-fold cross-validation were performed to evaluate model performance. Results showed that PF-LGBoost achieved an accuracy of 92.3% and an AUC of 0.942, outperforming competing models while maintaining a parameter size of only 1.2 MB and an inference time of 0.08 s per sample. These findings suggest that PF-LGBoost, with its remarkable lightweight advantage, enables efficient and deployable support for early fall risk stratification, rehabilitation monitoring, and individualized postoperative management in resource-constrained clinical settings for older adults after fracture surgery.
Language: English
Open Access status: hybrid
OpenAlex ID: W7213454530
FWCI:
Cites: 9
Cited by:
Related to: 10