TRUST-API: A Novel Deep-Learning-Based Adaptive Risk-Scoring Framework for Real-Time API Security in Digital Services

Application Programming Interfaces (APIs) are central to modern digital services, yet their distributed and dynamic nature creates security challenges that are not fully addressed by conventional binary or multiclass attack detection. This paper proposes TRUST-API, an adaptive deep-learning-based framework for real-time API risk assessment that transforms heterogeneous and chronologically evolving request behavior into a continuous and operationally interpretable security score. The framework integrates leakage-controlled preprocessing, behavioral feature engineering, chronological sequence construction, an Attention–BiLSTM temporal learner, explicit request-side risk components, calibrated nonlinear meta-risk estimation, and adaptive hybrid fusion. The experimental study uses 4282 labeled API requests for development and internal evaluation, together with a subsequent future-unlabeled request stream for distributional stability analysis. The Attention–BiLSTM converts posterior probabilities over Low-, Medium-, High-, and Critical-risk states into a continuous deep temporal-risk component, which is fused with contextual, header-morphology, authentication/session, and request-rate evidence to generate the final score, RTRUST. On the internal test set, TRUST-API achieved 99.518% accuracy, 99.689% Macro-F1, 99.760% balanced accuracy, 99.825% macro specificity, and a minimum class recall of 99.042%, with no Critical observations missed. Robustness was further assessed through ablation, sensitivity, additive-noise, missingness, moving-block bootstrap, chronological block analysis, and McNemar testing. The results demonstrate that TRUST-API provides accurate, stable, and risk-aware API security decisions while preserving an interpretable path from behavioral evidence to graded actions. External unlabeled observations are treated strictly as stability evidence rather than supervised performance validation. The resulting risk states support proportionate automated responses including normal access, enhanced monitoring, rate limitation, re-authentication, blocking, and security alert generation.

Authors

Institutions

Publication Details

Journal
Future Internet
Published
2026-11-18
DOI
https://doi.org/10.3390/fi18110637
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

TRUST-API: A Novel Deep-Learning-Based Adaptive Risk-Scoring Framework for Real-Time API Security in Digital Services

Mohammad O. Alhawarat, Yaseen Shehadeh, Ayoup Alsarhan
4 links
Future Internet
Network Security and Intrusion Detection
article

TRUST-API: A Novel Deep-Learning-Based Adaptive Risk-Scoring Framework for Real-Time API Security in Digital Services

Mohammad O. Alhawarat, Yaseen Shehadeh, Ayoup Alsarhan
article en

Abstract

Application Programming Interfaces (APIs) are central to modern digital services, yet their distributed and dynamic nature creates security challenges that are not fully addressed by conventional binary or multiclass attack detection. This paper proposes TRUST-API, an adaptive deep-learning-based framework for real-time API risk assessment that transforms heterogeneous and chronologically evolving request behavior into a continuous and operationally interpretable security score. The framework integrates leakage-controlled preprocessing, behavioral feature engineering, chronological sequence construction, an Attention–BiLSTM temporal learner, explicit request-side risk components, calibrated nonlinear meta-risk estimation, and adaptive hybrid fusion. The experimental study uses 4282 labeled API requests for development and internal evaluation, together with a subsequent future-unlabeled request stream for distributional stability analysis. The Attention–BiLSTM converts posterior probabilities over Low-, Medium-, High-, and Critical-risk states into a continuous deep temporal-risk component, which is fused with contextual, header-morphology, authentication/session, and request-rate evidence to generate the final score, RTRUST. On the internal test set, TRUST-API achieved 99.518% accuracy, 99.689% Macro-F1, 99.760% balanced accuracy, 99.825% macro specificity, and a minimum class recall of 99.042%, with no Critical observations missed. Robustness was further assessed through ablation, sensitivity, additive-noise, missingness, moving-block bootstrap, chronological block analysis, and McNemar testing. The results demonstrate that TRUST-API provides accurate, stable, and risk-aware API security decisions while preserving an interpretable path from behavioral evidence to graded actions. External unlabeled observations are treated strictly as stability evidence rather than supervised performance validation. The resulting risk states support proportionate automated responses including normal access, enhanced monitoring, rate limitation, re-authentication, blocking, and security alert generation.

Future InternetVol. 18(11)
Al-Ahliyya Amman University (JO), Hashemite University (JO)
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