As the global population continues to age, fall prevention has become a critical priority for preserving autonomy and quality of life among older adults. In response to the growing need for effective risk assessment tools, this study presents a methodological framework for the development of an interpretable and statistically robust fall risk score targeted at community-dwelling elderly individuals. The approach integrates multimodal data including clinical, stabilometric, and anamnestic information, and applies factor analysis to extract eight latent factors that capture the underlying structure of fall-related variables. Subsequently, Linear Discriminant Analysis (LDA) was employed to identify the combinations of latent factors most effective in classifying fallers and non-fallers. The best classification performance, considering a set of N = 26 older adults, was achieved using Factors 2, 5, 6, and 8, yielding an accuracy of 88%, sensitivity of 67%, and specificity of 95%. Protective factors against fall risk include mobility and balance capacity, younger age, and absence of respiratory problems or smoking habits. Conversely, factors contributing to increased fall risk include impaired balance and postural control, reduced muscle mass (e.g., sarcopenia), and physical conditions that affect gait quality without significantly affecting functional balance, such as in the early stages of Parkinson’s disease. These results enabled the formulation of a continuous composite fall risk score that combines predictive performance with clinical interpretability, thus overcoming a common limitation of current sensor-based and machine learning models. Despite its promising accuracy and high specificity, the tool exhibits moderate sensitivity, indicating that it should not replace a comprehensive clinical assessment during the physiatry visit. Instead, it is intended as a complementary support tool for clinicians, serving as an initial screening indicator to help identify people who may benefit from more in-depth evaluation. Future research will focus on expanding the dataset and incorporating additional sensing modalities, such as inertial measurement units, to enhance model generalizability and sensitivity.