Physically demanding tasks in the construction industry often lead to back-related musculoskeletal disorders among workers. Active back-support exoskeletons offer a promising ergonomic solution by augmenting human strength to reduce back strain. However, their use could introduce unintended consequences such as the risk of falls. Detecting fall risks is crucial for improving workers’ safety while using exoskeletons on construction sites. Foot plantar pressure provides insights into the pressure exerted beneath the feet, offering a means to evaluate stability and assess fall potential. This paper presents a machine learning framework to detect fall risk based on foot plantar pressure data collected during construction tasks involving exoskeletons, specifically carpentry framing tasks. Statistically significant differences in peak pressure between the left and right feet were used to distinguish between low and high-fall-risk scenarios during the tasks. Several classifiers including neural networks, ensemble, k-nearest neighbor, and support vector machine were employed to classify foot plantar pressure data into low and high fall risk categories. Results showed that support vector machines and ensemble methods outperformed other classifiers, achieving 63.7% accuracy with raw data and 71.9% accuracy with augmented data through jittering. This improvement highlights the effectiveness of data augmentation techniques. The study highlights the potential of machine learning techniques for real-time fall risk assessment of construction workers using active back-support exoskeletons. These findings motivate explorations of human-wearable robot interfaces to mitigate both musculoskeletal disorders and fall risks in occupational settings.