dataset_info:
features:
- name: id.resp_p
dtype: int64
- name: proto
dtype: string
- name: conn_state
dtype: string
- name: orig_pkts
dtype: int64
- name: orig_ip_bytes
dtype: int64
- name: resp_ip_bytes
dtype: int64
- name: label
dtype: string
splits:
- name: train
num_bytes: 132244
num_examples: 2370
download_size: 0
dataset_size: 132244
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
Aposemat IoT-23 - a Labeled Dataset with Malcious and Benign Iot Network Traffic
Homepage: https://www.stratosphereips.org/datasets-iot23
This dataset contains a subset of the data from 20 captures of Malcious network traffic and 3 captures from live Benign Traffic on Internet of Things (IoT) devices. Created by Sebastian Garcia, Agustin Parmisano, & Maria Jose Erquiaga at the Avast AIC laboratory with the funding of Avast Software, this dataset is one of the best in the field for Intrusion Detection Systems (IDS) for IoT Devices (Comparative Analysis of IoT Botnet Datasets).
The selection of the subset was determined by Aqeel Ahmed on Kaggle and contains 6 million samples. The Kaggle upload, nor this one, have employed data balancing. The Kaggle card does not contain methodology to understand what criteria was used to select these samples. If you want ensure best practice, use this dataset to mock-up processing the data into a model before using the full dataset with data balancing. This will require processing the 8GB of conn.log.labelled files.
This dataset only notes if the data is Malcious or Benign. The original dataset labels the type of malcious traffic aswell. This means this processing of the dataset is only suited for binary classification.
This dataset only contains the 6 most important columns: responder's port, transport layer protocol, connection state, number of packets sent by the originator, number of IP level bytes that the originator sent, number of IP level bytes that the responder sent, and the malicious or bengin label. These columns were determined from research done by Alani & Miri in "Towards an Explainable Universal Feature Set for IoT Intrusion Detection." They determined that these columns alone give 98% accuracy. This means a light weight model can contain these column alone and still have non-trivial results. This dataset only contains 2.5k rows, as the duplicates have been dropped.
Feature information:
All features originate from the Zeek processing performed by the dataset creators. See notes here for caviats for each column.
Expand for feature names, descriptions, and datatypes
Name: id.resp_p
Description: The responder’s port number.
Data type: int64 - uint64 in original
Name: proto
Description: The transport layer protocol of the connection.
Data type: string - enum(unknown_transport, tcp, udp, icmp). Only TCP and UDP in subset
Name: conn_state
Description: Value indicating connection state. (S0, S1, SF, REJ, S2, S3, RSTO, RSTR, RSTOS0, RSTRH, SH, SHR, OTH)
Data type: optional string
Name: orig_pkts
Description: Number of packets that the originator sent.
Data type: optional int64 - uint64 in original
Name: orig_ip_bytes
Description: Number of IP level bytes that the originator sent.
Data type: optional int64 - uint64 in original
Name: resp_ip_bytes
Description: Number of IP level bytes that the responder sent.
Data type: optional int64 - uint64 in original
Name: label
Description: Specifies if data point is benign or malicious.
Data type: string - enum(Malicious, Benign)
Citation
If you are using this dataset for your research, please reference it as “Sebastian Garcia, Agustin Parmisano, & Maria Jose Erquiaga. (2020). IoT-23: A labeled dataset with malicious and benign IoT network traffic (Version 1.0.0) [Data set]. Zenodo. http://doi.org/10.5281/zenodo.4743746”