Intrusion Detection In IoT Using Artificial Neural Networks On UNSW-15 Dataset

IoT devices are susceptible to numerous cyber-attacks due to its low power, low computational requirements and controlled environment that make it hard to implement authentication and cryptography in IoT devices. In this work we propose artificial neural network based threat detection for IoT to solve the authentication issues. We use supervised learning algorithm to detect the attacks and furthermore controller discards the commands after classifying it as threat. Proposed ANN consist of input, hidden and output layers. Input layer passes the data as signal to hidden layer where these signals are computed with the assigned weights and activation functions are used to transform an input to an output signal. Proposed technique is able to detect attacks effectively and timely decisions are taken to tackle the attacks. Proposed ANN approach achieves an average precision of 84% and less than %8 of average false positive rate in repeated 10-fold cross-validation. This reveals the robustness, precision and accuracy of proposed approach in large and heterogeneous dataset. Approach proposed in this work has the potential to considerably improve the utilization of intrusion detection systems.