AE-LSTM: Autoencoder with LSTM-Based Intrusion Detection in IoT

The acceleration in the field of the Internet of things had increased security problems, so we find ourselves in need of effective ways to protect IoT systems From intrusions. Recently Machine learning plays an active role in network security and detecting attacks. in this research, we propose a Machine learning method (AE-LSTM) for intrusion detection which uses Autoencoder with LSTM. Our method has 6-layer Autoencoder (AE) model with LSTM that is effective in anomaly detection. To avoid the bias in our model which occur from imbalanced data in the NSL-KDD dataset, we use Standard Scaler in our AE-LSTM model To delete the outliers from the input. AE-LSTM uses the best reconstruction function. It is critical in discovering whether network traffic is normal or abnormal. We use the NSL-KDD test dataset to evaluate our proposed model. Our Model achieved the highest accuracy over other methods with f1-score micro and weight at 98.69% and 98.70% for 5 classes in detection methods (Dos, Probe, R2L, U2R, Normal). Also, we evaluated it with two classes Malicious, Normal) with f1-score micro and weight at 98.78% and 98.78%.