The construction industry is among the most hazardous industries in the United States, associated with a high number of accidents. Workers' fatigue has been recognized as one of the four major causes of fatal incidents in this industry. Therefore, early identification of workers' fatigue in a project could support accident prevention. To this end, the objective of the present study is to develop a framework to detect workers' fatigue by examining their gait patterns measured by a three-axis accelerometer embedded in a smartphone. The application of accelerometer sensors in a smartphone is useful because it can record gait-pattern data at the construction site (not just limited just to a controlled environment). To achieve this objective, five construction workers were asked to participate in this study by recording their gait patterns before and after a fatigue-inducing exercise. Related time features were extracted and selected to train the classifier. Finally, supervised-learning algorithms [e.g., linear and nonlinear support vector machines (SVM)] were adopted to detect workers' fatigue in different working conditions. The study results indicate that workers' fatigue was detected at an accuracy of 87.93% and 82.75% using the linear and nonlinear SVMs, respectively. It is expected that these findings will provide useful guidelines for early prediction of physical fatigue and therefore enable project managers to make informed decisions in improving worker safety.