Development of Fixed Network Fault prediction model using deep learning in case of telecom

Fixed access networks play a very important role to provide quality telecommunication services in the current connected world. Ethio Telecom, Ethiopia's biggest telecommunications provider often finds it difficult to manage problems in its fixed-line network, providing broadband internet, voice calls, and business solutions. Conventional approaches to fault handling are mostly manual and reactive in nature, making it hard to manage the complexity, scale, and velocity needed in presentday telecom environments. This thesis suggests an innovative predictive model using deep learning to identify and classify faults within Telecom's fixed network. The study aims to build and train Artificial Neural Networks (ANN) and Bidirectional Long Short-Term Memory (Bi-LSTM) models with the help of historical customer complaint data. The data includes six months of running logs, divided into main fault types like copper faults, fiber optic faults, PSTN/voice service loss faults, and power faults. The models proposed are aimed at predicting network faults before they happen. This helps in conducting maintenance in advance and reducing service downtime. The Bi-LSTM model is suitable for time-series data and was more accurate and efficient than the existing models. The overall accuracy of the Bidirectional LSTM in a sequence model was as much as 98%. This means the model correctly predicted the type of fault in 98 out of 100 cases. Accuracy, which gauges the accuracy of the positive predictions for each class, averaged 0.98, while recall, which gauges the model's performance at predicting all relevant instances correctly, was 0.98 across all classes. The F1-score, which is the harmonic mean of recall and precision, was 0.98, reminding us about the harmony between the precision of prediction and the coverage. The results show that deep learning greatly improves predictive maintenance, lowers operational costs, increases network uptime, and supports Ethiopia's larger digital ambition under the Digital Ethiopia vision. The study adds to the body of research and offers useful information for telecom operators looking to improve the quality of service using intelligent automation and AI-powered fault management.