Shipped: AI Malware Detection Under Code Obfuscation

A machine-learning cybersecurity proof of concept evaluating how simulated obfuscation affects neural-network-based malware classification.

Using the BODMAS malware dataset and a TensorFlow/Keras classifier, the experiment evaluates performance across different simulated obfuscation conditions using accuracy, precision, recall, F1 score, and ROC AUC.

The project incorporates preprocessing, hyperparameter tuning, and cross-validation to study the relationship between feature transformations and malware detection performance.

Results are specific to the dataset and simulated transformations; they do not establish resistance to real-world malware packing or adaptive evasion attacks.