Detecting network intrusion using Probabilistic Neural Network

Intrusion detection plays an important role in solving network security problems. Artificial Neural Network (ANN) is one of the widely used intrusion detection techniques. However, many ANN-based methods are faced with unsatisfactory results and low detection precision. A new intrusion detection method by using Probabilistic Neural Network (PNN) is proposed. PNN divides inputs into two groups, normal and abnormal. Then different neurons are used to process these two different grouped inputs. Handling separately ensures that normal ones deviate from abnormal ones as far as possible, so as to obtain satisfactory detection results. PNN only needs one feed forward process and does not have any back propagation, thus greatly reducing the training time. Experimental results on KDDCUP99 dataset show that our PNN-based method yields average better performance than other well-knowns such as Decision tree, Naive Bayes and BPNN respecting precision, recall and F-value.