In the field of competitive sports, increasing training intensity is mostly used to improve sports and competition levels. However, after high-intensity training, the function of the body muscles will decrease, which is known as muscle fatigue. If there is a lack of reasonable control over exercise intensity, athletes may experience muscle fatigue and sports injuries. Based on this, this article takes the random forest algorithm of fiber optic sensors as the design basis and develops a fatigue detection system for sports competitions. This article first analyzes the relevant principles of fiber optic sensors, and based on their corresponding mode coupling theory, derives formulas for the output content of sensors. Through experiments, the advantages of fiber optic sensors are demonstrated. Then, the random forest algorithm was analyzed and improved, which belongs to the classic ensemble learning algorithm. The model used is universal, easy to understand, and not prone to overfitting. However, in terms of dynamic data classification, the performance is poor. The improved random forest algorithm not only has excellent detection performance, but also higher prediction accuracy than before. Finally, this article successfully developed a fatigue detection application system for sports competitions, which can detect the fatigue level of athletes in real time, adjust the intensity of exercise, and effectively avoid potential safety hazards during the exercise process.