logo
Search
Enhanced FOA with Dimensional Search Control and Memory-Based Strategy for Intrusion Detection Feature Selection

DOI: https://doi.org/jobasr

Abubakar Abdulkadir

Umar Iliyasu

Zaharraddeen Sani

Information

For Readers For Authors For Librarians

Announcement

A Publication of the Faculty of Physical Sciences, Federal University Dutsin-Ma, Dutsin-Ma, Katsina State, Nigeria.
Abstract
Feature selection plays a crucial role in improving the accuracy and efficiency of Network Intrusion Detection systems (NIDS) by reducing dataset dimensionality and eliminating redundant or irrelevant features that do not contribute meaningfully to classification outcomes. The Fruit Fly Optimization Algorithm (FOA) used for feature selection and its variants blindly search the solution space which leads to an imbalance between exploration and exploitation, reduce convergence speed and stuck at local optima. In this study, an enhanced feature selection algorithm based on fruit Fly Optimization Algorithm (FOA) is proposed to improve the balance between exploitation and exploration, faster convergence and avoid staganation at local optima. The enhanced version integrates two intelligent mechanisms Dimensional Search Control (DSC) and Memory-Based Strategy (MBS) which effectively guide and regulate the search process, enabling the algorithm to identify the most relevant features more efficiently.This study contribute by eliminating or reducing the high computational complexity faced by basic FOA and other FOA metaheuristic algorithm in feature selection and reduce the number of selected features as well as increase in the accuracy of the selected features. The proposed algorithm was implemented in Google Colab using Python programming language and evaluated using standard datasets NSL-KDD and CICID2017 against several well-known metaheuristic algorithms, including SCMWOA, BIFOA, ALO, and the basic FOA. The comparison was conducted using key performance metrics such as computational complexity (execution time and memory usage), number of selected features, classification accuracy, and fitness values .Experimental results demonstrate that the proposed enhanced FOA consistently outperformed the compared algorithms across all evaluation criteria as shown in section 4.1.5 fitness values of 99.9%,!00%,across the two datasets used, Section 4.1.4 accuracy values of 100% ,across the two datasets used, Section 4.1.3 28 and 18 number of selected features in NSL-KDD and CICID2017 datasets, Section 4.1.1 execution time of 854.45s and 4025.67s in NSL-KDD and CICID2017 respectively and Section 4.1.1 memory usage of 20.04 MB and 39.07 in NSL-KDD and CICID2017 respectively . Its superior efficiency, accuracy, and scalability make it highly suitable for deployment in modern Network Intrusion Detection System (NIDS) designs.
References
PDF