Wireless network throughput improvement using machine learning technique

• Main Author: Shyu, Zi Xun
• Other Authors: Law Choi Look
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Engineering; Info-communication
• Online Access: https://hdl.handle.net/10356/176331
• Institution: Nanyang Technological University

This project presents a viable method to optimize the wireless network by identifying the congestion level using machine learning techniques so that the user can switch to a network with lower congestion, to achieve a much higher network throughput. The data collected for machine learning is done with the help of a Raspberry Pi 4B and an Archer T2U wireless adapter. Data is collected with the help of Kismet, a wireless sniffer that can do Penetration Testing on wireless networks. Wireshark is used to help parse out the data from Kismet logging files. Jupyterlab with built-in Python was used to train the data set gathered, using the Enhanced Random Forest Model (ERFM) and Neural Network (NN). ERFM attained a much higher accuracy than NN, achieving 96.03% accuracy in predicting the traffic congestion in the network. The most important feature that surfaced to determine the Congestion level was the Length of the packet follow-up with Received Signal Strength Indicator. Further work can be done by gathering more data in Kismet, such as using Geographical Position System (GPS) coordinates to improve and identify the congestion level on a location basis.