MiniVHS-22: A Realistic Multi-Source Dataset for Intelligent Malware Detection Using Machine Learning

The rapid growth of malware and cyber threats demands intelligent methods for network traffic analysis. While machine learning (ML) techniques have shown high potential, their success is often limited by the quality and diversity of available datasets. Most existing network traffic datasets focus on a narrow range of malware types, reducing the generalizability of trained models. To address this gap, we present MiniVHS-22, an enhanced and compact version of the VHS-22 dataset. It is built by extracting flow-based parameters using a network probe and combining data from four public datasets with real-time samples from a malware monitoring website. This hybrid construction strategy significantly improves class diversity and real-world traffic representation. We evaluated MiniVHS-22 using seven ML algorithms. Among them, Random Forest, Decision Tree, and Multilayer Perceptron achieved over 99% accuracy in detecting malicious traffic. To optimize computational efficiency, we also applied dimensionality reduction techniques such as Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA). The novelty of this work lies in its real-time data integration, multi-source dataset fusion, and robust algorithmic evaluation. MiniVHS-22 provides a practical foundation for building scalable, adaptive, and intelligent threat detection systems for future cybersecurity challenges.