Determining the Number of Hidden Neurons in a Multi Layer Feed Forward Neural Network

摘要

A neural network intrusion detection system (IDS) can be effective against network attacks. However, their effectiveness can be reduced by changes in the neural network architecture. One problem is determining the number of hidden layer neurons. This can lead to reduced detection and high failure rates. This paper describes the affects of architecture on the performance of IDSs while finding a means to choose the proper architecture and number of hidden neurons. This method reduces the need for trial and error in determining the number of hidden layer neurons in a multi layer feed forward neural network IDS.