MCAD-CIC-3x1

Dataset Summary

MCAD-CIC-3x1 is a multi-source continual anomaly detection benchmark scenario for network intrusion detection. It combines three CIC-family source datasets into a three-task continual-learning scenario:

1. cicids2017
2. cicids2018
3. cicunsw

Each task corresponds to one consolidated source dataset. The benchmark is designed to evaluate continual anomaly detection methods under cross-source distribution shift.

The dataset contains 17,915,569 samples and has a reported 10.42% anomaly ratio in the test set.

Intended Use

This dataset is intended for research on:

• continual anomaly detection;
• continual learning for tabular data;
• network intrusion detection;
• robustness under distribution shift;
• forgetting and knowledge transfer across tabular anomaly detection datasets;
• benchmarking anomaly detectors under sequential task exposure.

The intended use is defensive machine learning research. The dataset should not be used to support offensive cybersecurity activity.

Dataset Sources

MCAD-CIC-3x1 is derived from the following source datasets:

• CIC-IDS2017: https://www.unb.ca/cic/datasets/ids-2017.html
• CIC-IDS2018: https://www.unb.ca/cic/datasets/ids-2018.html
• CIC-UNSW-NB15: https://www.unb.ca/cic/datasets/cic-unsw-nb15.html

The dataset contains aggregated flow-level tabular features derived from network traffic. Source and destination IP addresses and packet payloads are excluded during preprocessing.

Dataset Files

The repository contains the following files:

File	Description
data.csv	Main tabular dataset file.
orderings.json	Predefined task orderings for continual-learning evaluation.
croissant.json	Croissant metadata describing the dataset.
data_sampled.csv	Lightweight sampled subset for inspection.
data_sampled_readme.txt	Description of the sampled subset.

The sampled file is a balanced random sample of up to 1000 rows per task_id.

Dataset Structure

The main file is:

mcad-cic-3x1.csv

The dataset contains task metadata, binary labels, and numerical flow-level features.

Core Columns

Column	Type	Description
task_id	integer	Numeric identifier of the continual-learning task.
task_name	string	Name of the task/source dataset, e.g. cicids2017, cicids2018, or cicunsw.
task_split	string	Split assignment for the row.
label	integer	Binary anomaly label. 0 denotes benign/normal traffic and 1 denotes anomalous/attack traffic.

Feature Columns

The remaining columns are numerical network-flow features, including packet-count, byte-count, flag-count, duration, inter-arrival-time, and aggregate flow-statistics features. Representative examples include:

• Flow Duration
• Flow Bytes/s
• Flow Packets/s
• Total Fwd Packets
• Total Backward Packets
• Total Length of Fwd Packets
• Total Length of Bwd Packets
• Fwd Packet Length Mean
• Bwd Packet Length Mean
• Packet Length Mean
• Packet Length Std
• SYN Flag Count
• ACK Flag Count
• RST Flag Count
• Destination Port

For the complete schema, see croissant.json.

Task Orderings

The dataset provides six predefined orderings in orderings.json. These orderings define different continual-learning evaluation regimes over the same three tasks.

Ordering	Task sequence
curriculum_asc	cicunsw → cicids2017 → cicids2018
curriculum_desc	cicids2018 → cicids2017 → cicunsw
gen_paper_desc	cicids2018 → cicunsw → cicids2017
gen_paper_asc	cicids2017 → cicunsw → cicids2018
smooth_drift	cicids2018 → cicids2017 → cicunsw
abrupt_drift	cicids2017 → cicids2018 → cicunsw

These orderings are intended to expose complementary continual-learning dynamics, including curriculum-like adaptation, generalization-oriented ordering, smooth drift, and abrupt drift.

Dataset Creation

The details of dataset creation can be found in our NeurIPS submission paper.