SRM INSTITUTE OF SCIENCE AND TECHNOLOGY

Ramapuram, Chennai – 600 089

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

18CSP109L- MAJOR PROJECT

NETWORK TRAFFIC PREDICTION AND RESOURCE
MANAGEMENT

BATCH NUMBER : A6

Team Members Supervisor

VISHESH UPRETI NAME:
(RA2011003020528)
Ms. M. SNEHA PRIYA
SRIKANT DARIPA (ASSISTANT PROFESSOR SRM-RMP)
(RA2011003020535)

G. BHANU PRAKASH REDDY

(RA2011003020547)
 Agenda
• Abstract
• Scope and Motivation
• Introduction
• Literature Survey ( Table)
• Objective
• Problem Statement
• Proposed Work
• Architecture Diagram/Flow Diagram/Block Diagram
• Modules
• Module Description

• Software & Hardware Requirements

• References (Base paper hard copy to be submitted to the supervisor.)
• Way forward towards Outcome (Research Paper/Patent)

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

 ABSTRACT

• Utilizes a Recurrent Neural Network (RNN) architecture.

• Predicts heterogeneous multivariate Quality of Service (QoS)-aware
network traffic patterns.
• Dynamically allocates network resources based on QoS requirements
inferred from the predicted traffic patterns.
• Data is collected through Deep Packet Inspection (DPI), ensuring real-
world relevance and accuracy.

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

 Scope and Motivation

• Next-generation mobile networks prioritize connected intelligence.

• Demand for low latency, high bandwidth, and reliable connectivity is increasing.
• Real-time automated decision-making becomes necessary to handle the surge in
network traffic efficiently.
• Automated systems capable of dynamically allocating resources based on real-time
demands are essential.

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

 Introduction

• Next-generation networks like 5G and 6G are rapidly expanding.

• Efficient management of network resources becomes crucial to meet the
demands of modern applications and users.
• Automated orchestration solutions are essential for efficient resource
management.
• Existing architectures often lack detailed consideration of network slicing.
• Focuses on improving QoS while efficiently allocating resources based on
application requirements.

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

 Objective
• Develop a predictive framework for heterogeneous multivariate Quality of Service
(QoS)-aware network traffic.
• Leverage a Recurrent Neural Network (RNN) integrated with Bidirectional Long
Short-Term Memory (BLSTM) architecture.
• Enable dynamic allocation of network resources to optimize performance and
mitigate congestion.
• Consider metrics such as precision, execution time, and energy consumption.

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

 Problem Statement
1. Network Traffic Prediction: Traffic in a network with p Quality
of Service (QoS) classes is observed over K time slots. The
objective is to predict the traffic volume for each QoS class at
the next time stamp, minimizing the loss function that
measures the disparity between predicted and actual traffic
volumes.
2. Dynamic Resource Allocation: During network congestion,
fair allocation of bandwidth is crucial, especially for delay-
sensitive traffic. Therefore, the next step involves dynamically
assigning network resources to each QoS class based on
predicted traffic volumes and the priority of each class.

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

 Proposed Work

The proposed framework leverages a hybrid Recurrent Neural Network (RNN)

incorporating Bidirectional Long Short-Term Memory (BLSTM) units to
dynamically predict QoS-aware network traffic. Through this fusion, it
achieves exceptional predictive capabilities, accurately forecasting network
traffic for over 13 hours with an average accuracy of 97.68%. This model
showcases the potential of advanced neural network architectures in
effectively addressing the challenges of next-generation mobile networks,
offering valuable insights for proactive resource allocation and management.

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

Architecture Diagram

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

 Modules
• Data Preprocessing
• Dataset Reduction By Correlation Analysis
• Support Vector Machine Regressor
• Evaluation Metrics

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

 Module 1: Data Preprocessing
Data normalization ensures balanced feature influence in machine
learning by scaling numerical data. Min-max normalization, common
for preserving data relationships, confines values within [0,1].
Handling missing data, vital for dataset integrity, includes deletion,
manual filling, or replacing with constants or means. Using feature
means, like maximum likelihood, maintains dataset
representativeness. Adaptations may employ class-specific means.
Overall, normalization aids algorithm efficacy, with min-max scaling
and strategic handling of missing data tailored to dataset nuances and
task demands.

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

 Module 2: Dataset Reduction
Correlation analysis identifies linear relationships
between variables. By visualizing histograms and scatter
plots, closely linked variables are pinpointed, aiding in
redundancy removal. High correlation, depicted by
Pearson coefficients exceeding 0.8 or below -0.8, signals
redundancy. This ensures streamlined datasets for
efficient data processing techniques.

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

 Module 3 : SVM Regressor
• Support Vector Machines (SVMs) excel in sparse, noisy data
for classification, extending to regression with Support Vector
Regression (SVR). SVR adapts SVM principles for continuous
response variables, optimizing hyperplanes through training
algorithms. While SVMs handle minimal data in the margin
well, SVR's performance decreases with more data points in
the margin.

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

 Module 4 : Evaluation Metrics
• For time series data evaluation, various metrics are utilized.
MAE measures average absolute differences between
predicted and actual values, RMSE emphasizes large errors,
MAPE assesses percentage differences, and MASE compares
model accuracy against a naive historical model. These
metrics offer diverse insights into predictive accuracy,
ensuring comprehensive assessment of machine learning
tools' performance on time series datasets.

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

 Software & Hardware Requirements
Hardware Requirements:

• RAM - 4 GB

• Processor - 1.2 GHz dual core (minimum)

• hard drive - 10 GB (minimum)

Software Requirements:

• Browser compatibility - Microsoft internet explorer, Microsoft edge, Mozilla Firefox, Google Chrome,
Safari

• Operating System - Windows 2012 R2 or above

General Standards:

• Must be easy and intuitive to use for the target audience.

• Must function in a logical manner for the target audience.

Development Technology, Programming Language, and Web Server Software:

• Google collaboratary

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

 References
[1] U. Cisco, ‘‘Cisco annual internet report (2018–2023) white paper,’’ Cisco, San Jose, CA, USA, White Paper
C11, 2020, vol. 10, no. 1, pp. 1–35. [Online]. Available: https://www.cisco.com/c/en/us/solutions/
collateral/executive-perspectives/annual-internet-report/white-paper-c11- 741490.html
[2] N. H. Mahmood et al., ‘‘White paper on critical and massive machine type communication towards 6G,’’
2020, arXiv:2004.14146.
[3] Y. Guo, Q. Duan, and S. Wang, ‘‘Service orchestration for integrating edge computing and 5G network: State
of the art and challenges,’’ in Proc. IEEE World Congr. Services (SERVICES), Oct. 2020, pp. 55–60.
[4] H. Chergui, A. Ksentini, L. Blanco, and C. Verikoukis, ‘‘Toward zero-touch management and orchestration of
massive deployment of network slices in 6G,’’ IEEE Wireless Commun., vol. 29, no. 1, pp. 86–93, Feb. 2022.
[5] J. Kaur, M. A. Khan, M. Iftikhar, M. Imran, and Q. E. Ul Haq, ‘‘Machine learning techniques for 5G and
beyond,’’ IEEE Access, vol. 9, pp. 23472–23488, 2021.
[6] Zero-Touch Network and Service Management (ZSM); Reference Architecture, document ETSI GS ZSM 002,
Group Specification (GS) ETSI GS ZSM, 2019.
[7] Experiential Networked Intelligence (ENI) Requirements, document ENI, version 2.1. 1, ETSI GS ETSI, Sophia
Antipolis, France, 2019.
[8] K. Velasquez, D. P. Abreu, M. Curado, and E. Monteiro, ‘‘Resource orchestration in 5G and beyond:
Challenges and opportunities,’’ Comput. Commun., vol. 192, pp. 311–315, Aug. 2022.
[9] Distributed Artificial Intelligence-Driven Open & Programmable Architecture for 6G Networks, ADROIT6G,
Athens, Greece, 2023.
[10] J. Zhang and N. Ansari, ‘‘On assuring end-to-end QoE in next generation networks: Challenges and a
possible

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

 Outcome

The framework proposed herein utilizes RNN-BLSTM neural networks to

identify, classify, and predict heterogeneous network traffic. Predictions for
each Quality of Service (QoS) class enable optimized resource allocation,
particularly in dynamically assigning links based on traffic requirements. With
an impressive average prediction accuracy of 97.68%, exceeding existing
methods, it enables proactive resource allocation, mitigating future
congestion for sensitive internet traffic. Comparative results showcase its
superiority over fair bandwidth allocation schemes. Future research aims to
validate the framework using network simulators, enhancing its applicability
and robustness.

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING