An Advanced AI and Machine Learning Framework for IoT Driven Data Science, Edge Intelligence, Real Time Pattern Recognition, and Autonomous Decision Making
Journal of Intelligent Decision Making and Information Science · 10.10.2026 · 3 (17s) · 970-998
DOI: 10.17762/jidmis.v3.5452 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Intelligent Decision Making and Information Science, online erschienen am 10.10.2026. Verfasst von Sourav Kumar Biswas. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Background
Industrial Internet of Things (IIoT) deployments generate high-rate, heterogeneous sensor streams whose value depends on converting raw signals into timely, reliable decisions close to the data source. Integrated evaluations that link edge-side feature engineering, pattern recognition, drift handling, resource cost, and downstream decision quality under one statistically rigorous protocol remain scarce.
Methods
We propose EI-Frame, a seven-layer IoT–edge–cloud framework combining per-device commissioning baselines, a 60-dimensional time–spectral feature engine, gradient-boosted pattern recognition, a label-free drift tracker, an int8-quantisable lightweight head, and a shielded reinforcement-learning maintenance agent. EI-Frame was evaluated in a fully reproducible, physics-inspired simulation of 24 heterogeneous rotating-machine nodes (7,200 labelled four-channel windows; five operating states) using leave-devices-out validation over 10 independent repetitions, Friedman and Wilcoxon signed-rank tests with Holm correction, paired effect sizes (Cohen’s dz), and 95% confidence intervals.
Results
EI-Frame achieved a macro-F1 of 0.974 ± 0.007 (95% CI [0.969, 0.979]), outperforming gradient boosting without device normalisation (0.953 ± 0.014; Δ = 0.021, Holm-adjusted p = .008, dz = 2.02) and all other baselines (Friedman χ²(4) = 37.36, p < .001, Kendall’s W = 0.93). A device-normalised max-|z| novelty screen reached an AUROC of 0.978, exceeding Isolation Forest variants. Under progressive load, thermal, and noise drift, label-free baseline tracking preserved late-stream macro-F1 at 0.879 versus 0.610 for the static model (dz = 5.13). Post-training int8 quantisation of the lightweight head retained 99.92% prediction agreement in a 9.7 kB footprint with 0.15 ms median inference latency. Condition-based decision policies eliminated nearly all failures relative to run-to-failure and improved episode return over tuned periodic maintenance by 13.28 units; however, tabular Q-learning offered no advantage over a well-specified threshold rule.
Conclusions
In this controlled setting, device-level normalisation and drift-aware baselining—not model complexity—were the dominant sources of robustness, and learned decision policies added little beyond transparent rules. Validation on physical testbeds and public benchmarks is required before operational claims are made.
Zitieren
APA
Biswas, S. K. (2026). An Advanced AI and Machine Learning Framework for IoT Driven Data Science, Edge Intelligence, Real Time Pattern Recognition, and Autonomous Decision Making. Journal of Intelligent Decision Making and Information Science, 3(17s), 970-998. https://doi.org/10.17762/jidmis.v3.5452
Vancouver
Biswas SK. An Advanced AI and Machine Learning Framework for IoT Driven Data Science, Edge Intelligence, Real Time Pattern Recognition, and Autonomous Decision Making. Journal of Intelligent Decision Making and Information Science. 2026;3(17s):970-998. doi:10.17762/jidmis.v3.5452
BibTeX
@article{Biswas2026,
author = {Biswas, Sourav Kumar},
title = {An Advanced AI and Machine Learning Framework for IoT Driven Data Science, Edge Intelligence, Real Time Pattern Recognition, and Autonomous Decision Making},
journal = {Journal of Intelligent Decision Making and Information Science},
year = {2026},
volume = {3},
number = {17s},
pages = {970--998},
doi = {10.17762/jidmis.v3.5452}
}
EUVIMED