Machine Learning Approach for Automatic Fault Detection and Diagnosis in Cellular Networks

Abstract

The capability for a network to self heal itself is a promising feature for future cellular networks. An essential function to achieve self healing is the ability to determine when a network is operating outside of normal state, and perhaps identify potential causes. This paper focuses on applying the supervised machine learning approach to detect fault symptoms and identify the cause. Our method utilizes referenced signal received power (RSRP) reported by users over a certain period of time to detect operational anomaly in a base station. We notice that certain faults at a base station create noticeable change in the RSRP readings and recognizable electromagnetic radiation pattern around the base station. To achieve fault analysis, we develop a framework that differentiates normal and abnormal operations under changing environment to avoid unnecessary fault alarms. Once abnormal operation is detected, the framework uses a supervised machine learning system to classify the detected fault. We develop convolutional neural network and random forest to test the fault classification. We show that both machine learning systems offer high accuracy.