Malware Prediction Using Tabular Deep Learning Models

As technology progresses, malware evolves, becoming increasingly perilous and posing a significant challenge in combating cybercriminals. With the abundance of massive data on vulnerabilities, Deep learning techniques present a chance to further boost data and system security. This paper introduces a deep neural network model that automatically generates embedding layers for each categorical feature. Its foundation lies primarily in training neural oblivious decision ensembles and TabNet model on malware data, benefiting from both end-to-end gradient-based optimisation and the power of multi-layer hierarchical representation learning. These deep architectures possess the capacity to learn numerous parameters and identify patterns within large-scale datasets. The proposed models were evaluated using the Microsoft malware prediction dataset, which includes nine million labelled subjects and 83 features. This work marks one of the early attempts to utilise deep tabular architectures for malware prediction. The experimental results demonstrate the model's effectiveness, achieving an accuracy of 66.1% and AUC of 72.8%.