Testing Deep Learning Methods to Predict Ransowmare Activity from Hybrid Analysis

Alexander M. Veach (Eastern Michigan University, USA)
Munther Abualkibash (Eastern Michigan University, USA)

International Journal of Informatics, Information System and Computer Engineering (INJIISCOM)
Vol. 7 No. 1 (2026): INJIISCOM: VOLUME 7, ISSUE 1, JUNE 2026 (ONLINE FIRST)

This study applies deep learning methods to predict ransomware using hybrid analysis samples. To understand current detection methods, prior research was analyzed, guiding the creation of an experiment that tests a model built from ransomware hybrid analysis. A training dataset of over 500 samples, encompassing 38 ransomware families and benign Windows programs, was utilized. The resulting model was subsequently evaluated against a testing dataset containing novel ransomware families not represented during training, revealing a notable performance decrease. Comparing these findings with existing literature highlights potential flaws in how artificial intelligence models are tested and reported. Consequently, this paper advocates for more complex prediction methods and alternative strategies to guarantee models maintain external effectiveness.