Traffic optimization in modern mobile ad networks
Kolotilin, Grigorii (2025)
2025
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2025052415613
https://urn.fi/URN:NBN:fi:amk-2025052415613
Tiivistelmä
Modern mobile ad networks face multiple challenges in their business. One of the key problems is the amount of traffic. Billions of ad requests every day should be handled properly, fast, and precisely. In this context, the traffic optimization problem started to be one of the most important to solve.
The main objective of this master's thesis was to develop a framework for traffic optimization, which helps to optimize ad request flow and helps to predict whether each ad request should be served or throttled. In the first phase of this research, it was explored how the mobile ad networks handle traffic and what the space for improvement is.
The second phase was about different approaches for traffic optimization, how these approaches are implemented, and which approach suits which case.
The third phase was about the implementation process, which started as a simple baseline model and finished with the fully custom production-grade LightGBM model. The developed model was successfully tested in the product environment. The model performance was evaluated with certain metrics such as recall, precision, confusion matrix, etc.
The developed traffic optimization framework showed significant results, which were compared with the theoretically possible maximum. In addition to it, the framework could be utilized for some different tasks, such as a smart throttling mechanism for incidents.
The main objective of this master's thesis was to develop a framework for traffic optimization, which helps to optimize ad request flow and helps to predict whether each ad request should be served or throttled. In the first phase of this research, it was explored how the mobile ad networks handle traffic and what the space for improvement is.
The second phase was about different approaches for traffic optimization, how these approaches are implemented, and which approach suits which case.
The third phase was about the implementation process, which started as a simple baseline model and finished with the fully custom production-grade LightGBM model. The developed model was successfully tested in the product environment. The model performance was evaluated with certain metrics such as recall, precision, confusion matrix, etc.
The developed traffic optimization framework showed significant results, which were compared with the theoretically possible maximum. In addition to it, the framework could be utilized for some different tasks, such as a smart throttling mechanism for incidents.