Performance Analysis of Ensemble Machine Learning Algorithms in PDF Malware Detection

Because of its adaptability, independence from platforms, and capacity to embed various kinds of content, PDF is one of the most widely used document file formats. Throughout time, PDF has emerged as a common attack vector for disseminating malware and infiltrating computer systems. Despite having incredibly high recall rates, existing signature-based defense systems are easily evaded by malicious PDF files since they soon become outdated and ineffective against malicious attacks. Research investigations showed that machine learning approaches offer effective anti-malware detection solutions. The CIC-Evasive-PDFMal2022, a newly published dataset, is utilized in our research study to test seven machine learning classifiers, including both traditional and ensemble approaches, in terms of accuracy, precision, recall, f1-score, ROC-AVC curve and operation time. The result suggests ANN, XGBoost, and RFC consistently demonstrated high levels of accuracy, with a rate of over 98% in all cases concluding that the ensemble techniques predict better while working with large feature sets. Furthermore, performance of XGBoost can be considered the best it has higher accuracy with smaller operation time.