Falls remain a frequent and debilitating problem worldwide.1 They are considered a health problem and a unique senile syndrome.2 In the USA, as many as one in four older adults fall yearly,3 and falls contribute to accidental deaths among older adults aged ≥75 years in other countries.4

Like in other developed countries, falls are the leading cause of trauma in older adults in Singapore, accounting for 85 % of all older adults with trauma seen in the emergency department. With one of the most rapidly aging populations worldwide, falls among older adults are an urgent public health challenge in Singapore.5 In a 2019 survey of national living conditions in Japan,6 falls were the fourth leading cause of older adults needing care after dementia, cerebrovascular disease (stroke), and age-related frailty, accounting for 12.5 %. The survey also indicated that three times more older adults die from falls and collisions than from road traffic accidents. Not only older adults but also their families are aware of the dangers of falls6. However, there is no single cause of a fall; it results from an individual's exposure to multiple risk factors.2,7

Falls are a multifactorial problem that can be attributed to intrinsic (behavioral, physical, and cognitive) and environmental causes.8,9 Therefore, multifactorial risk assessment has proven more effective than unifactorial risk assessment at capturing the complex nature of falls among community-dwelling older adults.10,11,12 A systematic review by Vandervelde12 comprehensively explored fall prevention among community-dwelling older adults, showing that multifactorial fall prevention programs appear effective for individuals with a previous fall history. Costello and Edelstein13 also noted that exercise alone effectively reduced falls, and fall prevention programs should include muscle strengthening and balance improvement. Toraman and Yildirim7 highlighted that medication and intrinsic factors, such as age and diseases, that affect the functions needed to maintain balance are often related to fall risk.

Previous studies have focused on the efficacy of multidisciplinary fall prevention clinics. Pre- and post-intervention studies with a control group have previously examined multifactorial preventions involving exercise training and clinical treatment.14,15,16,17 Most of these studies used traditional statistical control groups to determine trait values, and their assumptions were validated based on physicians’ inferences about the factors of interest, perhaps limiting other factors. With the problem of physical decline with aging attracting the attention of public health and geriatric departments worldwide, studies have examined this issue, finding that functional decline in older adults is closely associated with problems such as insufficient physical activity and frequent sedentariness, which endanger their ability to move independently and consequently increasing their mortality.18,19,20

Taiwan currently has a treatment-oriented healthcare and medical system, with the implementation of preventive medicine and health promotion at the front end relatively lacking. Therefore, studies have examined physical fitness and fall assessment using questionnaires and statistical methods.18,21,22 Statistically, while multiple regression has been used for analysis, assessing the empirical factors selected by experts is easier. For example, Soriano et al.2 identified predisposing risk factors, including the sensory and central nervous systems. They highlighted that physical examinations should focus on gait and balance and that the analysis of multiple risk factors is challenging. In addition, multifactor detection for many individuals has not yet been explored in comprehensive clinical screenings.

The advantage of artificial intelligence is its rapid accumulation of experience. In machine learning, logistic regression, random forest, and Relief algorithms are often used to explore eigenvalues.23,24 However, few studies have explored using different machine learning algorithms in analyzing physical fitness for fall risk. Using logistic regression, Toraman and Yildirim7 determined that agility and dynamic balance are relevant to falling risk.

Traditional analytic approaches, such as logistic regression, are commonly used to examine the influence of one or more risk factors on falls simultaneously.25 If several measurements are taken, the amount of data that must be analyzed increases. To facilitate machine learning, data dimensionality reduction can help reduce the calculation load and time.26 However, fall prevention for community-dwelling older adults involves multiple risk factors and intervention measures, making data analysis challenging.12 The data are also “big,” with numerous variables and dimensions. While the number of variables should be reduced to facilitate data analysis, basic information should not be lost. Principal component analysis (PCA) is commonly used for dimensionality reduction and has several clinical applications. Studies on fall risk have used PCA to reduce data dimensionality to obtain optimum solutions regarding daily-life gait quality27 and gait dynamics.28 However, feature selection methods have not yet been explored for analyzing clinical data or initially screening community-dwelling older adults.

Traditional approaches, such as multiple regression, are commonly used to simultaneously analyze the influences of one or more risk factors on falls. However, the primary prevention of falls among community-dwelling older adults involves multiple risk factors and intervention measures, and data analysis is challenging due to the large number of variables. In this study, we use statistical analysis and machine learning to explore multiple factors derived from the participants’ basic information, exercise plans, and community health promotion activities and identify those that might lead to falls. This study's conceptual framework is shown in Fig. 1.