ML-Based Optical Fibre Fault Detection in Smart Surveillance and Traffic Systems

It is evident that the intense transformation in the smart city structure has produced a demand for more optical fibre networks to facilitate the systems' speedy communication for instance traffic control, surveillance, as well as loT devices. Due to the nature of the optical fibre networks being very susceptible, and the slightest break or a bend can result in a major breakdown of operation; then, the ability to quickly identify the fault as well as rectify it is important in maintaining the efficiency of the systems. In this work, we propose a detailed workflow for fibre optic fault detection and classification using machine learning. We employ LightGBM, XGBoost, CatBoost, and AdaBoost machine learning models, along with OTDR data to categorize fault types. The process we adopt comprises enhancing the raw data to capture more of the signals quality before analyzing the data using these models for fault detection. Of all the models LightGBM was the best performing as it recorded an accuracy of 98.12 % thereby making it to be the best model for this task. The use of key performance metrics such as accuracy, precision, recall, and F1-score along with confusion matrices, ROC curves on the graphs was done in order to measure the performance of the models accurately. Based on the performance of these models, a rational strategy in developing an intelligent solution for maintaining the operability and efficiency of smart city fibre optic networks is achieved.