An Efficient Botnet Detection Using Machine Learning and Deep Learning

Bots are the major threats that are used to compromise network. For botnet detection, various machine learning and deep learning models are proposed. But, the research lacks in efficient pre-processing of the available dataset. While removing null values from the dataset, to handle duplicate network flow, a label ‘Weight’ is added which reduces the dataset enormously and helps in efficient botnet detection too. Further, to handle class-imbalance problem, ADASYN and SMOTE methods are applied separately, and their impact is checked. In this paper, three machine learning namely, Gaussian Naive Bayes, KNN, and linear regression methods as well as three deep learning methods, namely CNN, LSTM, and MLP are applied on pre-processed and balanced CTU-13 dataset. Later, various performance metrics evaluate these models to identify bots. The results show that pre-processed and balanced dataset gives accurate results without overfitting.