Research on a multidimensional quantitative assessment method for the walking ability of elderly people based on machine learning and multi-strategy balanced optimizers

Abstract As people age, older adults inevitably experience declines in stability, balance, endurance, and muscle strength, leading to issues with walking ability. To address the limitations of single-function walking ability tests, this study proposes a comprehensive quantitative method for assessing walking ability in older adults based on multi-modal physiological data, including heart rate, surface electromyography, and inertial measurements. First, experimental results from three classic tests—the 10MWT, 6MWT, and TUG—were collected, quantified, graded, and weighted to derive the T-10-6 index, T-10-6 showed excellent agreement with expert consensus (weighted κ = 0.87, 95% CI 0.83–0.89; Spearman ρ = 0.91, 95% CI 0.68–1.00). Subsequently, a decision tree was used to construct a basic feature assessment system (BFAS), and correlation weight optimization system (CWOS) and a multi-strategy improved balanced optimizer (MSEO) were introduced to perform collaborative optimization of weights and thresholds, ultimately forming the MSEO-TO-CW model. Compared to the T-10-6 benchmark, which integrates the three classic tests, the consistency of MSEO-TO-CW improved from 77.14% to 91.43% (95% CI: 84.87–97.99%) using BFAS, and its evaluation results outperformed the classic Holden scale at 82.86% (95% CI: 74.03–91.69%), with a consistency rate of 86.67% with clinical expert diagnoses, and external-dataset validation further showed 80% agreement between MSEO-TO-CW and the benchmark. Experiments demonstrated that this method performs optimally in single-feature, single-modality, and multi-modality fusion scenarios, providing an efficient and reliable new tool for the objective assessment of lower limb function and precise rehabilitation training in the elderly.