Efficient Throughput Degradation Prediction in Telco Networks using Anomaly Detection

Proactive anomaly detection of the Key Performance Indicators (KPIs) in a telecom network is important for consistent end-user experience. In this paper, we present an application of state-of-the-art deep learning based multivariate time series Long Short Term Memory (LSTM) model which can forecast KPIs based on historical data. Our approach is able to efficiently predict KPIs by learning patterns from the time series data along with the seasonality behaviour. We also predict anomalies on this forecasted data by the application of Isolation Forest, that is tuned with the contamination hyperparameter to indicates the severity of the anomalies. We evaluate our approach on real data from a LTE network and observed good results for forecasting the throughput KPI and predicting anomalies.