Enhancing Network Fault Prediction with Survival Analysis and Machine Learning: the Telenium Software Experience

Telenium is an advanced network management software designed to keep networks performing optimally. The software records timestamps of alarms for four types: critical, major, minor and NSA, with critical alarms suggesting the biggest negative impact. On average 5% of devices show at least one critical alarm each day. This makes prioritization of inspection, maintenance check and repair difficult. By using historical activity data, this paper presents a hybrid approach combining survival analysis and machine learning model to enhance fault predictions. It shares best practices in experimental design, sampling, feature generation, feature selection and model evaluation.