Prediction of traffic volume in BTS sites using deep learning

Payload and Throughput are the most essential parameters in determining the congestion as well as performance of the network. Predicting these parameters in advance is a proactive approach for LTE network management and planning. Mostly ARIMA which are linear models have been applied in predicting f...

Full description

Saved in:
Bibliographic Details
Main Authors: Sagar, Chilakala Jithendra, Gupta, Hardik, Jadon, Jitendra Singh, Arora, Neha, Kumar, Sachindra
Format: Conference Proceeding
Language:English
Subjects:
Computer architecture
Deep learning
GRU
Hidden Markov models
KPI
Logic gates
LSTM
Microprocessors
Neural Networks
Payloads
Predictive models
RNN
Telecom Networks
Throughput
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Payload and Throughput are the most essential parameters in determining the congestion as well as performance of the network. Predicting these parameters in advance is a proactive approach for LTE network management and planning. Mostly ARIMA which are linear models have been applied in predicting future trends. However in this paper we propose special kind of Recurrent neural Network architectures mainly, LSTM and GRU, as these networks are capable in remembering longer dependency, learning temporal patterns in a long sequence of random length. These networks are used for predicting the payload and throughput trend with the help of past Key performance indicator reports. In the proposed work KPI data of past 16 days from Nokia Networks Pvt Ltd is obtained for research purpose. In the proposed work, RNN architectures achieved state of the art results on live data set from Nokia Networks producing promising results in predicting payload and throughput.
ISSN:2688-769X
DOI:10.1109/SPIN48934.2020.9070897