Towards Reliable 6G: Intelligent Trust Assessment with Hybrid Learning ^†

Abstract

Sixth-generation (6G) networks will operate with pervasive autonomy and minimal centralised control, imposing stringent requirements on security and trust. This short communication presents a hybrid trust evaluation approach that combines fuzzy inference for uncertainty management, bidirectional long short-term memory (BiLSTM) networks for temporal prediction, and blockchain for immutable verification. The pipeline first maps multi-source interaction and context metrics into linguistic trust values via fuzzy rules, then leverages BiLSTM to anticipate trust fluctuations under dynamic conditions, and finally anchors trust updates on a permissioned blockchain to ensure integrity and traceability. Using CIC-IoT2023, the proposed approach attains high accuracy and F1-score while reducing Execution Time (ET) and energy demands relative to a recent spatial-temporal trust model for 6G IoT. Results indicate that jointly addressing uncertainty, temporal evolution, and ledger-backed validation yields stable trust trajectories suitable for resource-constrained devices. The study outlines a practical path toward explainable, adaptive, and tamper-resistant trust management for 6G ecosystems.

1. Introduction

The convergence of massive IoT, edge intelligence, and near-real-time orchestration in 6G expands the attack surface and elevates the importance of trust among heterogeneous nodes. Conventional trust mechanisms are typically centralised or static, leaving them ill-suited for rapidly varying traffic, mobility patterns, and adversarial behaviours such as spoofing or routing manipulation. Three gaps are recurrent: (i) robust handling of uncertainty in noisy metrics, (ii) sensitivity to the temporal evolution of trust, and (iii) protection of trust records against tampering. This work introduces a hybrid trust evaluation approach that integrates fuzzy inference, BiLSTM-based temporal learning, and blockchain-backed verification. The design objective is an interpretable, adaptive, and verifiable mechanism that can be deployed on constrained devices while maintaining reliable performance.

2. Related Work

Fuzzy-based trust models provide an interpretable aggregation of noisy and heterogeneous indicators, improving stability under uncertainty but typically lacking explicit temporal learning [1]. Learning-driven approaches enhance detection and prediction; yet, many omit sequential dependencies or do not guarantee the integrity of trust records across domains [2,3]. Blockchain-oriented schemes address tamper resistance and auditability for cross-administrative settings in IoT/6G [4], but they do not natively model time-varying trust signals. Among recent contributions, Ma et al. proposed a spatial–temporal trust model for 6G IoT using clustering and moving-average smoothing [5]. In contrast, the present work unifies fuzzy uncertainty handling with learnt temporal dynamics (BiLSTM) and permissioned-ledger validation to concurrently address interpretability, adaptivity, and integrity within a concise pipeline.

3. Materials and Methods

3.1. Data and Preprocessing

Experiments use CIC-IoT2023 [6], which provides traffic and context-level features that reflect large-scale IoT activity. For each node i at time t, we form a normalised feature vector $x_i\left(t\right)\in {[0,1]}^d$ from reliability (delivery/loss), QoS (latency/bandwidth/service reliability), and context (mobility/energy/environment). Missing values are imputed by per-feature means; infinities are capped to the maximum finite value. Data is split 80/20 for training/testing.

3.2. Fuzzy Inference

A compact rule base aggregates indicators into a preliminary trust score. Inputs are fuzzified into Low/Medium/High using triangular membership functions. Representative rules include:

IF (delivery is High) AND (latency is Low) THEN trust is High; IF (loss is High) OR (mobility is High) THEN trust is Low.

Defuzzification uses the centroid method, producing a continuous preliminary trust value ${\tau }_f\left(t\right)\in [0,1]$.

3.3. Temporal Prediction (BiLSTM)

To capture sequential dependencies, a BiLSTM consumes recent windows of preliminary trust and auxiliary indicators to predict near-term trust $\widehat{\tau }(t+1)$. Training uses Adam (fixed learning rate), early stopping, and minibatches sized for edge-grade GPUs/CPUs. The bidirectional structure provides context from both past and (in training) future indices, improving robustness to abrupt transitions.

3.4. Blockchain Validation

Each finalised trust update is hashed and appended to a permissioned blockchain via a smart contract that enforces role-based submission and prevents unauthorised overwrites. This layer delivers data integrity, non-repudiation, and auditable provenance. The final decision fuses components as a convex combination:

$${\tau }_{\mathrm{final}}\left(t\right)=\alpha {\tau }_f\left(t\right)+\beta \widehat{\tau }\left(t\right)+\gamma {\tau }_{\mathrm{ledger}}\left(t\right),\alpha ,\beta ,\gamma \ge 0,\alpha +\beta +\gamma =1,$$

(1)

with weights tuned to minimise validation error while stabilising variance.

4. Results

4.1. Setup and Metrics

We assess accuracy, precision, recall, F1-score, mean absolute error (MAE), ET, Energy Consumption (EC), and trust-variance stability. Baselines consider the spatial-temporal trust model by Ma et al. [5]. All models are implemented in a uniform environment with identical train/test splits.

4.2. Quantitative Outcomes

The proposed approach attains high classification quality on CIC-IoT2023, with F1-scores near 0.98 and low variance in trust trajectories. Relative to Ma et al. [5], it yields higher precision/recall while reducing ET and EC, indicating suitability for constrained deployments.

Table 1. The proposed hybrid approach improves both predictive metrics and efficiency with the reference method.

Method	Precision	Recall	F1-Score	ET (s)	EC (J)	TV
Proposed hybrid model	0.985	0.978	0.982	1.25	0.87	0.0025
Spatial–temporal model ([5])	0.812	0.791	0.801	2.65	1.98	0.0156

Column headers summarise the following: ET = Execution time (seconds), EC = Energy consumption (joules), TV = Trust variance.

summarises the main indicators.

4.3. Visualisation

Figure 1 depicts the end-to-end workflow (inputs, fuzzy layer, BiLSTM prediction, and ledger anchoring). Figure 2 illustrates representative trust time series on the test split, showing smooth yet responsive adaptation under fluctuating conditions.

5. Discussion and Conclusions

The results demonstrate that coupling fuzzy reasoning with BiLSTM-based temporal learning and blockchain-backed validation provides complementary benefits: interpretability under uncertainty, responsiveness to time-varying behaviour, and tamper resistance for multi-stakeholder settings. Compared with the spatial-temporal baseline proposed by Ma et al. [5], the approach improves predictive quality and efficiency while preserving the stability of trust trajectories. These characteristics are pertinent for large-scale 6G deployments that must operate with limited energy budgets and require auditable trust management across domains. Future work will investigate adaptive weight tuning and broader cross-domain validation, as well as latency-aware ledger configurations for ultra-dense scenarios.