HTTP Botnet Detection Using Adaptive Learning Rate Multilayer Feed-Forward Neural Network

Botnets have become a rampant platform for malicious attacks, which poses a significant threat to internet security. The recent botnets have begun using common protocols such as HTTP which makes it even harder to distinguish their communication patterns. Most of the HTTP bot communications are based on TCP connections. In this work some TCP related features have been identified for the detection of HTTP botnets. With these features a Multi-Layer Feed Forward Neural Network training model using Bold Driver Back-propagation learning algorithm is created. The algorithm has the advantage of dynamically changing the learning rate parameter during weight updation process. Using this approach, Spyeye and Zeus botnets are efficiently identified. A comparison of the actively trained neural network model with a C4.5 Decision Tree, Random Forest and Radial Basis Function indicated that the actively learned neural network model has better identification accuracy with less false positives.