Journal of Artificial Intelligence, Applications and Innovations

Bayesian deep learning for collaborative spectrum sensing in 6G mmWave communication systems

Mahdi Nouri , Elaheh Karimpour Fard , Sima Sobhi-Givi , Hamid Behroozi ^*

In this paper, we present an innovative approach for spectrum sensing in mmWave massive multiple-input multiple-output (MIMO) systems by leveraging a hierarchical Bayesian model integrated with Stacked Denoising Autoencoders (SDAEs). The proposed method significantly improves the accuracy of spectrum sensing by effectively estimating the unknown covariance of the CSCG noise. By incorporating Bayesian deep learning principles into the model, a robust estimation of noise characteristics is achieved, enhancing the detection performance of weak signals amidst noise with unknown covariance. The use of Collaborative Deep Learning (CDL) allows for adaptive learning of complex noise patterns and provides a more accurate characterization of the spectrum. Extensive simulations are conducted to validate the effectiveness of the proposed CDL-based spectrum sensing method in mmWave massive MIMO systems. The simulations compare the performance of the Bayesian deep learning approach against traditional spectrum sensing techniques and state-of-the-art methods. The results demonstrate superior performance in terms of detection accuracy, noise robustness, and computational efficiency. Detailed performance metrics and analysis are provided to showcase the practical advantages of integrating CDL with SDAEs for spectrum sensing in real-world mmWave scenarios.

Bridging Data Silos for Enhanced Predictive Maintenance: A Federated Learning Framework

Khalil Jahani iD , Behzad Moshiri ^* iD , Babak Khalaj iD

This paper presents a secure federated learning framework for industrial predictive maintenance that addresses data silos arising from privacy, competition, and regulatory constraints. The proposed approach combines PCA-based anomaly screening, LSTM-based self-encryption, client credibility scoring, CKKS homomorphic aggregation, and differential privacy to enable collaborative model training without sharing raw sensor data. Experiments on real-world vibration and acoustic datasets demonstrate that the proposed framework achieves failure detection performance close to centralized training, outperforming conventional federated baselines such as FedAvg and FedProx. Specifically, the model attains an average F1-score of 89.4 ± 0.5% and a remaining useful life prediction error (MAPE) of 8.7 ± 0.6%, while reducing communication overhead. Practical deployment considerations, including cryptographic overhead and hardware constraints, are critically discussed, highlighting trade-offs between security, efficiency, and scalability. The proposed framework provides a viable and privacy-preserving solution for predictive maintenance in Industry 4.0 environments.

A Novel AI-Driven Framework for Predicting Fragmentation and Cost-Oriented Design Optimization Using a Hybrid XGBoost–NSGA-II Approach

Hassan Hosseinzadeh ^* , Gholamreza Nowrouzi iD , Elham Rafieenia

Modeling and optimizing the blasting process in mining has always been one of the fundamental challenges in mining engineering due to its multidimensional and nonlinear nature. This study aimed to develop a hybrid framework based on artificial intelligence for predicting rock fragmentation and simultaneously optimizing blasting cost and quality. In the first stage, the XGBoost machine learning algorithm was employed to model and predict the mean fragmentation size (P₅₀) based on blasting design parameters. The results indicated that the proposed model could accurately capture complex relationships among variables, achieving R² values of 0.97 for the training dataset and 0.92 for the testing dataset, thus demonstrating remarkable predictive performance. Variable importance analysis revealed that the specific charge (q) and the burden distance (B were the most influential factors controlling fragmentation, while the rock mass quality (T) also played a decisive role in altering the fracture mechanism. In the next step, the NSGA-II evolutionary algorithm was applied for multi-objective optimization between blasting costs and the P₅₀ index. The optimization outcomes generated a set of Pareto solutions, allowing engineers to flexibly select blasting patterns in accordance with either economic or operational priorities. The findings demonstrate that the proposed hybrid framework not only enhances prediction accuracy compared to empirical models but also provides a powerful data-driven decision-making tool for blast design. By introducing an integrated approach based on XGBoost and NSGA-II, this study makes a significant contribution to the blasting engineering literature and paves the way for developing sustainable and efficient blast designs in various mining operations.

PeerReli-AI: AI-Enhanced Guidance for Boosting Peer Feedback Reliability

Seyede Fatemeh Noorani ^* , Hossein Morovvati , Hassan Morovvati , Amir Hushang TajFar

Advances in artificial intelligence (AI) offer new opportunities to deliver scalable and personalized formative feedback in higher education, yet robust empirical evidence of their effectiveness remains limited. Although peer feedback can promote active learning, reflective thinking, and evaluative judgment, inconsistencies and inaccuracies in student assessments often undermine its impact.

This study proposes PeerReli-AI, an AI-supported peer feedback framework that integrates automated quiz administration, peer evaluation, and AI-generated formative guidance. A messaging-bot infrastructure was used to deliver quizzes and manage anonymous peer feedback, while the Gemini large language model generated context-aware feedback on students’ peer evaluations. The framework was implemented in an undergraduate Data Structures and Algorithms course across multiple iterative assessment cycles.

Results from Hotelling’s T² tests and linear mixed-effects models demonstrate a significant reduction in discrepancies between peer and instructor evaluations over successive iterations when AI guidance was provided. Survey findings further indicate that students perceived the AI-generated feedback as engaging, useful, and supportive of their learning.

Overall, the findings suggest that PeerReli-AI enhances the reliability and quality of peer feedback and offers a scalable approach for improving formative assessment practices in large higher education classrooms.

A Hybrid Pattern Extraction and Confidence-Weighted Volume Framework for Machine Learning-Based Forex Price Forecasting

Maysam Yazdanpanahi ^* , Morteza Zahedi iD , Mohammad Mehdi Hosseini iD

Since its inception, the Forex market has attracted various market players, including central banks, financial institutions, and even academic researchers, due to its high volatility and potential profitability following the technical analysis saying that "history tends to repeat itself", RWBCPE captures the historical patterns to anticipate future changes (up or down) within certain possible boundaries to avoid signals that lack confidence thresholds. RWBCPE underwent testing on the EUR/USD currency pair in the context of three different simple machine learning models: Random Forests, Support Vector Machines, and Long Short-Term Memory networks to demonstrate its strength and efficiency, even using simple, conventional machine learning and deep learning models. The framework achieved a higher accuracy than the traditional technical indicators, such as RSI and MACD. To evaluate the practical feasibility of the framework, we conducted a 25-year training set from 2000 to 2025 and a 1-year backtest from 2025 to 2026 with an initial balance of $10,000. The RWBCPE-driven strategy achieved 65.96% accuracy and 67.29% precision, growing the account balance to $34,844.75. The system also recorded a profit ratio of 2.17 and a maximum drawdown of 4.6%, demonstrating strong predictive performance and low-risk trading through confidence-weighted position size.

STEM-Net: An Explainable Multi-Task Hybrid Architecture Combining Tabular Transformers, Gradient-Boosted Trees and Ensemble Learning for the Simultaneous Prediction of Eight Psychological and Academic Problems in High-School Students

AbdolRahim Papi iD , Maseud Rahgozar ^* iD , Habibollah Arasteh Rad iD , Arshia Badi iD , Mohammad Hossein Rezvani iD

Psychological and academic problems in students rarely occur in isolation: depression, anxiety, low self-esteem and addiction proneness tend to appear together, and modelling each of them separately wastes the signal they share. This study introduces the STEM-Net framework. Data were collected from 1,713 high-school students using eight validated questionnaires; after synthetic minority oversampling (SMOTE) the set grew to 5,139 records, and labels were coded at several risk levels based on the official cut-off points of each instrument. This multi-task hybrid architecture predicts all eight problems simultaneously, each at the risk levels defined by its own instrument. The base layer contains five heterogeneous learners in three branches: a multi-task tabular transformer whose attention body is shared across the eight targets, gradient-boosted tree models trained per target, and a pre-trained network. Before fine-tuning, the body of the deep branch is pre-trained on roughly eighty-eight thousand public respondents to two psychometric instruments, and the outputs of the five learners are combined through an automatic choice between stacking and simple averaging. To prevent item-to-score leakage, the items of the target scale are removed from the feature space of that task. On the test set STEM-Net reaches a macro-average accuracy of 0.9277, an F1 of 0.9034 and a ROC-AUC of 0.9844. A four-method explainability analysis shows that in every one of the eight tasks the most informative source is another one of the scales, and that addiction proneness and suicidal ideation are mutually predictive. Building on these features, a web-based screening system with a short questionnaire was designed that reports, for each student, the predicted risk level together with the features that most influenced it.

Detection of Epileptic Spikes using the Wavelet-RLS Hybrid Method

Mohsen Shafieirad ^* , Nastaran Salehoun , Maryam Songhorzadeh , Zohreh Aarabi

https://doi.org/10.61838/jaiai.2.1.1

Epilepsy is a neurological disorder characterized by recurrent seizures caused by abnormal neural activity in the brain. Accurate detection of epileptic spikes in EEG signals is crucial for effective treatment. In this study, we propose a hybrid method that combines wavelet transform with Recursive Least Squares (RLS) adaptive filtering to enhance EEG signals by removing artifacts. The method's performance was evaluated using the Signal-to-Noise Ratio (SNR) and Mean Squared Error (MSE), demonstrating superior artifact removal performance compared to traditional approaches. Wavelet-based statistical features were then extracted and classified using a Probabilistic Neural Network (PNN), which is recognized for its high accuracy and computational efficiency. The proposed method achieved a classification accuracy of approximately 91.86% on real-world EEG data.

Modeling Ankle Joint Dynamics Based on Inertial Sensor Data Using Machine and Deep Learning

Alireza Naeimi-Sadigh ^* iD , Abdollah Shirazi

https://doi.org/10.61838/jaiai.2.1.5

In recent years, human motion analysis using inertial measurement units (IMUs) has emerged as a lightweight, low-cost, and portable alternative to advanced biomechanical equipment such as optical motion capture systems or force plates. IMUs enable the recording of motion data in real-world and daily settings, making them powerful tools for medical, sports, and rehabilitation applications. This study proposes a compact and intelligent framework for modeling ankle joint dynamics using only a single IMU mounted on the right ankle, with the dual aim of classifying movement activities and estimating selected dynamic parameters. Data were obtained from the publicly available HuGaDB dataset, comprising accelerometer and gyroscope signals sampled at 100 Hz. The signals were segmented using sliding windows, and 36 statistical features were extracted from each segment. Significant differences between classes were examined using the Kruskal–Wallis test, and the most relevant features were selected with the SelectKBest algorithm. Three modeling approaches were evaluated: Gaussian Process Regression, Random Forest, and a hybrid Convolutional Neural Network with Long Short-Term Memory. The CNN+LSTM model achieved the highest classification accuracy at 96.41%, outperforming the other models, while Random Forest reached 90.5% accuracy using the top 10 selected features. GPR produced satisfactory results for continuous parameter estimation. Training time, resource consumption, and model size were also assessed. The findings demonstrate that high accuracy in human motion analysis can be achieved using data from a single IMU. The proposed framework holds promise for clinical, sports, and rehabilitation contexts, as well as for developing lightweight and personalized wearable systems.

Journal of Artificial Intelligence, Applications and Innovations

About the Journal

Current Issue

Articles

Bridging Data Silos for Enhanced Predictive Maintenance: A Federated Learning Framework

PeerReli-AI: AI-Enhanced Guidance for Boosting Peer Feedback Reliability

STEM-Net: An Explainable Multi-Task Hybrid Architecture Combining Tabular Transformers, Gradient-Boosted Trees and Ensemble Learning for the Simultaneous Prediction of Eight Psychological and Academic Problems in High-School Students

A Hybrid Pattern Extraction and Confidence-Weighted Volume Framework for Machine Learning-Based Forex Price Forecasting

Bayesian deep learning for collaborative spectrum sensing in 6G mmWave communication systems

A Novel AI-Driven Framework for Predicting Fragmentation and Cost-Oriented Design Optimization Using a Hybrid XGBoost–NSGA-II Approach

Language

Issues

Journal Statistics Summary

Average time from submission until

Indexing & Abstracting

Time from submission to

Make a Submission

Editorial Board

Keywords

Bayesian deep learning for collaborative spectrum sensing in 6G mmWave communication systems

Bridging Data Silos for Enhanced Predictive Maintenance: A Federated Learning Framework

A Novel AI-Driven Framework for Predicting Fragmentation and Cost-Oriented Design Optimization Using a Hybrid XGBoost–NSGA-II Approach

PeerReli-AI: AI-Enhanced Guidance for Boosting Peer Feedback Reliability

A Hybrid Pattern Extraction and Confidence-Weighted Volume Framework for Machine Learning-Based Forex Price Forecasting

STEM-Net: An Explainable Multi-Task Hybrid Architecture Combining Tabular Transformers, Gradient-Boosted Trees and Ensemble Learning for the Simultaneous Prediction of Eight Psychological and Academic Problems in High-School Students

Detection of Epileptic Spikes using the Wavelet-RLS Hybrid Method

Modeling Ankle Joint Dynamics Based on Inertial Sensor Data Using Machine and Deep Learning