Hybrid Approach for Improving Intrusion Detection Based on Deep Learning and Machine Learning Techniques

Abstract

An intrusion detection system works to recognize the attacks using either the signature or signature-less method. The signature-less method suffers from a lot of false alarms that affect accuracy and recall. Commonly used IDS (intrusion detection system) Dataset experiences imbalance which causes a high false alarms rate. Nowadays CNN (convolution neural network) excels in image and computer vision. Using CNN in IDS is promising. The paper proposes a hybrid approach between CNN and ML (SVM, KNN). CNN is efficiently utilized to get important features from the dataset. Then ML used to classify the data. Using the hybrid approaches to benefit from the advantage of machine learning (high accuracy, Low false alarms) and Deep learning which deal with a large amount of data and reduce the number of feature of the dataset (feature extraction). In this paper we used 10% of KDDcup1999 dataset. The experimental results showed enhancement in the detection accuracy to 99.3 and reduction in losses to 0.03.