attack_error_latencies

time_series_latency

This repository includes all the codes used for the paper:

Detection Latencies of Anomaly Detectors: An Overlooked Perspective?

1) Step1. We collect two public datasets: ROSPaCe, and Arancino. The first is a dataset for intrusion detection composed by monitoring an embedded system on normal behavior and under attack. The second is specific for error detection and it represents an embedded system in an Internet of Things setting. Both datasets are time series-based and comprise a set of sequences of variable length.

2) Step2. We preprocess each dataset to extract sequences and shuffle their order. Note that we do not alter the order of data points within each sequence.

3) Step3. We select 5 representative detectors from the state-of-the-art solutions reviewed and organized into algorithms for unordered data points and algorithms for ordered data points.

4) Step4. We split the two datasets into train, validation, and test sets, using 60% of sequences for the training, 30% for the test, and 10% for validation. We train the selected detectors on each dataset.

5) Step5. We evaluate the models using state-of-the-art metrics and our proposed metrics.

Check README file inside the repositories in the getting_started.zip and all_experiments.zip.