Abstract: In order to improve the recognition ability of abnormal data in optical fiber communication networks, an abnormal data detection algorithm based on the optimization of entropy objective function was proposed. Firstly, the attributes of the data samples were classified, and the radius of the neighborhood interval was selected based on the abnormal data feature density index. Secondly, the clustering degree of the big data with high-order statistics was iterated and the optimization of the feature extraction parameters was completed. Finally, the optimal value of the entropy objective function was calculated according to the sample attribute probability, and it was used to complete the abnormal data detection. The experiment tested 1 000 sets of communication data. The results show that the average detection accuracy of this algorithm is about 95.7%, and its data fusion rate, detection time-consuming and average false detection rate are better than that of the two traditional methods. It can be seen that this algorithm has the advantages of high accuracy, fast convergence and low false detection rate, which has the certain application value.