Abstract
The field of information and computer security is rapidly developing in today’s world as the number of security risks is continuously being explored every day. The moment a new software or a product is launched in the market, a new exploit or vulnerability is exposed and exploited by the attackers or malicious users for different motives. Many attacks are distributed in nature and carried out by botnets that cause widespread disruption of network activity by carrying out DDoS (Distributed Denial of Service) attacks, email spamming, click fraud, information and identity theft, virtual deceit and distributed resource usage for cryptocurrency mining. Botnet detection is still an active area of research as no single technique is available that can detect the entire ecosystem of a botnet like Neris, Rbot, and Virut. They tend to have different configurations and heavily armored by malware writers to evade detection systems by employing sophisticated evasion techniques. This report provides a detailed overview of a botnet and its characteristics and the existing work that is done in the domain of botnet detection. The study aims to evaluate the preprocessing techniques like variance thresholding and one-hot encoding to clean the botnet dataset and feature selection technique like filter, wrapper and embedded method to boost the machine learning model performance. This study addresses the dataset imbalance issues through techniques like undersampling, oversampling, ensemble learning and gradient boosting by using random forest, decision tree, AdaBoost and XGBoost. Lastly, the optimal model is then trained and tested on the dataset of different attacks to study its performance.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have