Author(s): Nisha

Abstract: This study presents a deep learning-based threat detection framework for cybersecurity applications, focusing on accurately identifying malicious network activity. The methodology begins with the collection of network traffic data using the CICIDS2017 dataset, which includes realistic benign traffic and contemporary attack scenarios, addressing limitations of previous datasets such as insufficient diversity and incomplete metadata. Pre-processing ensures high-quality data by removing noise, handling missing values, encoding categorical features, normalizing numerical attributes, and addressing class imbalances. Exploratory Data Analysis (EDA) is performed to uncover patterns, outliers, and feature correlations, guiding effective model selection. Both baseline machine learning models—Logistic Regression, Random Forest, and XG-Boost—and deep learning models, including a Fully Connected Feedforward Neural Network (FNN) and a 1D Convolutional Neural Network (1D-CNN), are implemented and evaluated. Comparative analysis shows that while classical models perform reasonably well, deep learning models, particularly the 1D-CNN, achieve superior performance, with an accuracy of 97.5%, high precision, recall, and AUC metrics. The results demonstrate the framework’s ability to capture complex feature interactions and sequential patterns in network traffic, providing robust, reliable, and generalizable detection of cyber threats. This study highlights the effectiveness of deep learning approaches in enhancing real-world intrusion detection systems.

DOI: 10.33545/26633582.2025.v7.i2b.206

Pages: 117-126 | Views: 656 | Downloads: 453

Download Full Article: Click Here