Intrusion Detection Using Machine Learning And Deep Learning for IoT (FYP)

It is an implementation of our research work. We implemented DT, RF and Bi-LSTM on ARP_MITM and SSDP FLOOD datasets. For RF and DT, concatenated dataset was divided into three ratios: 70%:30%, 80%:20%, and 90%:10% for training and testing respectively. Both models were trained on separate datasets as well with the same ratios. For Bi-LSTM, concatenated dataset was used with different number of layers. All the models performs very well and had an accuracy over 99% but Random Forest outperform all the models. The front-end of our prototype, demonstrate the dataset upload process, which includes both malicious and normal packets. Upon uploading the dataset, the back-end model conducts predictions on the packets, accurately categorizing them as normal or malicious. The front-end then presents the performance metrics and confusion matrix, providing valuable insights into the model's effectiveness.

https://ieeexplore.ieee.org/document/10099152

1. The files are notebook files so jupyter notebook IDE or Dataspell IDE is required.
2. You can install the dataset from this link. https://www.kaggle.com/datasets/ymirsky/network-attack-dataset-kitsune
3. The front-end prototype is made using streamlit so you can run it on pycharm. Open the project in pycharm, goto terminal, run the command "streamlit run main.py"

It's my FYP with my FYP partner Sarim Amir https://github.com/SarimAmir and my supervisor Misbah Anwer and Yumna Iftikhar.

We only trained our model on 19GB dataset out of 64GB because of the limited resources. You can built a more powerful intrusion detection model by using the whole dataset. You can also apply different algorithms and different techniques to increase the accuracy.