Explainable IoT intrusion detection via bio-inspired encoding and interpretable AI models

Abstract We introduce an explainable Internet of Things (IoT) intrusion detection framework built on a bio-inspired Amino Acid encoding of network flows. The flow features are first transformed with the Yeo–Johnson transformation, the most informative ones are selected, and their values are encoded as Amino Acid sequences, which are then summarized by ten protein structural properties (e.g., $$\beta $$ -Turn, Isoelectric Point and Molecular Weight). In this way every flow is described by the same compact set of ten properties, whatever the feature set of the original dataset. The models are explained with SHAP, LIME and ELI5, complemented by Captum Integrated Gradients and Alibi anchors for the neural models, and the attributions are linked back to the raw traffic features through measured correlations. On a balanced 50,000-flow subset of CICIoT2023 that covers all 33 attack classes, most of the 12 classical and 5 deep models reach 98.4–98.5% accuracy on a held-out test set (MCC up to 0.97). The recurrent neural network (RNN), used for the in-depth explainability study, reaches 98.5% accuracy with a false-alarm rate of 0.02%, has fewer than 15,000 parameters and needs about 0.1 ms per flow on a CPU. An ablation experiment shows that encoding the selected features keeps detection close to the full 39-feature raw representation (MCC 0.970 against 0.979), while giving a fixed ten-property input that can be explained. All explanation methods agree that $$\beta $$ -Turn and Isoelectric Point drive most decisions, and both properties correlate strongly with the packet count and packet size of the flows.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-15
DOI
https://doi.org/10.1038/s41598-026-79957-2
Primary Topic
Network Security and Intrusion Detection
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Explainable IoT intrusion detection via bio-inspired encoding and interpretable AI models

Muhammad Kazim, Mustafa Kaiiali, Stefan Kühn, Mujeeb Ur Rehman et al.
Scientific Reports
Network Security and Intrusion Detection
article

Explainable IoT intrusion detection via bio-inspired encoding and interpretable AI models

Muhammad Kazim, Mustafa Kaiiali, Stefan Kühn, Mujeeb Ur Rehman, Thaer AL Ibaisi
article en

Abstract

Abstract We introduce an explainable Internet of Things (IoT) intrusion detection framework built on a bio-inspired Amino Acid encoding of network flows. The flow features are first transformed with the Yeo–Johnson transformation, the most informative ones are selected, and their values are encoded as Amino Acid sequences, which are then summarized by ten protein structural properties (e.g., $$\beta $$ -Turn, Isoelectric Point and Molecular Weight). In this way every flow is described by the same compact set of ten properties, whatever the feature set of the original dataset. The models are explained with SHAP, LIME and ELI5, complemented by Captum Integrated Gradients and Alibi anchors for the neural models, and the attributions are linked back to the raw traffic features through measured correlations. On a balanced 50,000-flow subset of CICIoT2023 that covers all 33 attack classes, most of the 12 classical and 5 deep models reach 98.4–98.5% accuracy on a held-out test set (MCC up to 0.97). The recurrent neural network (RNN), used for the in-depth explainability study, reaches 98.5% accuracy with a false-alarm rate of 0.02%, has fewer than 15,000 parameters and needs about 0.1 ms per flow on a CPU. An ablation experiment shows that encoding the selected features keeps detection close to the full 39-feature raw representation (MCC 0.970 against 0.979), while giving a fixed ten-property input that can be explained. All explanation methods agree that $$\beta $$ -Turn and Isoelectric Point drive most decisions, and both properties correlate strongly with the packet count and packet size of the flows.

Scientific Reports
University of Tartu (EE), De Montfort University (GB)
Network Security and Intrusion Detection
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon
Explainable IoT intrusion detection via bio-inspired encoding and interpretable AI models — Muhammad Kazim, Mustafa Kaiiali, et al. · Scientific Reports (2026) | TGRS Research Map | TGRS