A harmony search–driven hyperparameter optimization framework for CNN-based intrusion detection
Year: 2026
Type: article
Abstract: Convolutional Neural Networks (CNN) have been applied extensively in Intrusion Detection Systems (IDS) due to their capability of automatically learning hierarchical features representations of network traffic. However, the detection capabilities of CNN depend significantly on the quality of features and appropriate setting of hyperparameters configurations. In order to tackle these problems, this research proposes CNN hyperparameters optimization using harmony search (HS) optimization technique for developing an optimized one-dimensional convolutional neural network (1D-CNN) based IDS. Correlation matrix analysis is used in the preprocessing stage to select the features corresponding to the class of interest and remove redundancy among the highly correlated input features. Afterwards, the proposed method uses harmony search (HS) optimization algorithm, which is inspired by musical improvisation process, for optimization of CNN hyperparameters through exploration and exploitation of the search space. The proposed technique has been tested on NSL-KDD dataset using four classes of traffic: Normal, DoS, Others and Probe. Experimental results on test data set with 22,544 records show an overall accuracy of 97.50%. This model is able to yield results of the weighted Precision, weighted Recall and weighted F1-Score of approximately 97%. This reveals the power that can be unleashed by integrating data pre-processing approaches such as feature engineering based on correlation, Quantile Transformation, Class-weight and Hyperparameter optimization using Harmony Search technique in 1D-CNN based intrusion detection systems.
Source: Discover Networks
Authors: Shruti, Prashant Pranav
Institution: Birla Institute of Technology, Mesra
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid:
OpenAlex ID: W7234594088
