Dealing with High Cardinality of Network Management System Data for Machine-Learning-Based Alarm Classification

MOTIVATION

Accurate and prompt alarm classification is indispensable for network operators to manage failures effectively

Machine learning-based solutions are receiving significant attention, but they are often constrained by the high cardinality of the dataset extracted from the Network Management System (NMS). This is because most of the features are categorical, with hundreds of unique labels

To avoid the curse of dimensionality and achieve optimal performance from ML models, the high cardinality of data must be dealt efficiently