A Comparison of Classical and Quantum Neural Network Algorithms for Malware Detection

In recent years, malicious software has risen in both number and effectiveness of attacks, particularly due to the increased utilization of machine learning and even quantum computing by cybercriminals. This research addresses the disparity between these rapidly progressing technologies and the lack of advanced cybersecurity systems in nearly half of all American small businesses by creating accessible malware detection software on Google Colaboratory, a cloud-based software hosting service. Neural networks (NN), a common machine learning model for cybersecurity tasks, were created with both classical and simulated quantum computing using Python libraries, starter code, and an Android malware dataset. The average F1 score of accuracy of the classical NN, 0.6928, was slightly higher than the average F1 score of the quantum NN, 0.6406. However, a two-sample t-test found that the true means of both populations of F1 scores were statistically equal, indicating no difference in accuracy between classical and simulated quantum neural networks at this time. Differences in resource efficiency could not accurately be assessed, because although the quantum neural network was simulated on a classical computer and had a much higher runtime, a quantum neural network running on a quantum computer can be significantly faster than a classical neural network running on a classical computer for certain types of problems; this is because of unique quantum properties like superposition and entanglement that allow complex calculations to be performed much more efficiently. This research does find, however, that features related to the services or transactions of Android applications had higher correlations with malicious/benign software status, the change in runtime of a classical NN on Google Colaboratory was roughly linear over a large range of sample sizes, and two iterations of text segments and comments within the software improved the usability of the tool. Overall, the study develops a methodology for creating and evaluating open-source quantum and classical machine learning cybersecurity software for small businesses.