A Hybrid Combination of LS-SVM and KPCA with Bat Algorithm for Intrusion Detection

Abstract

Background: Intrusion Detection System refers to a kind of software which is designed for securing a network or an information system by alerting the administrators automatically when someone is trying to compromise the overall system through malicious activities. An intrusion is a process of causing a system to enter into an insecure state. Thus, an intruder is a person or an attacker who attempts to violate the security measures by interrupting the integrity and confidentiality of a system. Objective: Today, security threads are becoming big issues when working with high speed network connections, one solution is the intrusion detection system which promises to provide network security. This paper proposed Least Square Support Vector Machine (LS-SVM) based on Bat Algorithm (BA) for intrusion detection. Methods: The proposed techniques have divided into two phases. In the first phase, KPCA is utilized as a preprocessing of LS-SVM to decrease the dimension of feature vectors and abbreviate preparing time with a specific end goal to decrease the noise caused by feature contrasts and enhance the implementation of LSSVM. In the second phase, least square support vector machine with a BA is applied for the classification of detection. BA utilizes programmed zooming to adjust investigation and abuse amid the hunting procedure. Finally, as per the ideal feature subset, the feature weights and the parameters of LS-SVM are optimized at the same time. Results: The proposed algorithm named is Kernel principal component analysis based least square support vector machine with bat algorithm (KPCA-BA-LS-SVM). To show the adequacy of proposed method, the tests are completed on KDD 99 dataset which is viewed as an accepted benchmark for assessing the execution of intrusions detection. Furthermore, our proposed hybridization method gets a sensible execution regarding precision and efficiency. Conclusion: A BA model with a novel hybrid KPCA-LS-SVM was proposed in this paper for intrusion detection system. The parameters for LS-SVM classifier are chosen with the usage of BA, the necessary features of intrusion detection data was separated using KPCA and the multi-layer SVM classifier. This classifier checks whether any activity encounters an attack. The N-RBF supports the Gaussian kernel work by measuring the preparation time and forwarding the LS-SVM classifiers execution. As future work, several algorithms should be developed by combing other available classification methods with kernel systems resulting in efficient planning of examination and online intrusion detection.