Scopus Research — Murtada Dohan

This study introduces an innovative framework called Wavelet-Based Bidirectional Long Short-Term Memory (W-BiLSTM) designed to enhance spectral and energy efficiency in millimeter-wave (mmWave) large scale multiple-input multiple-output (LS-MIMO) systems. The approach combines wavelet transforms for efficient signal decomposition and noise reduction with BiLSTM networks to extract robust features and optimize hybrid beamforming. By integrating these techniques, the W-BiLSTM framework improves channel estimation precision and beamforming performance, tackling key issues related to energy usage and spectral efficiency in advanced wireless communication systems. Comprehensive simulations reveal that the W-BiLSTM framework outperforms existing methods such as Dual-Deep-Network (DDN), Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), and Autoencoders. Notably, it achieves spectral efficiency improvements of up to 13.5 bps/Hz and energy efficiency gains of up to 50 bits/Joule, alongside a substantial reduction in bit error rate (BER) across various signal-to-noise ratios (SNRs). These findings underscore the framework’s capability to address the demanding needs of 5G and future wireless technologies, offering a pathway toward more efficient and dependable communication systems. To improve reproducibility and benchmarking, the suggested framework is also assessed using the publicly accessible DeepMIMO dataset (Scenario O1), demonstrating its resilience under standardized mmWave conditions. © This article is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License. License details: https://creativecommons.org/licenses/by-sa/4.0/

Precise commodity price prediction is essential for financial markets, supply chain management, and risk management. In this research, Support Vector Machine (SVM), Long Short-Term Memory (LSTM), and Random Forest (RF) models are applied with the BIST dataset to forecast gold, silver, and nickel prices. The highest accuracy was attained by the LSTM model with an R2 of 0.94 and MAPE of 1.8%, surpassing SVM (MAPE 2.5%) and RF (MAPE 3.7%). SVM had robust short-term predictions, enhancing forecast accuracy by 1 0-1 5 % compared to linear models, while RF showed greater variance, being more suitable to identify volatility in the market. The latency of real-time predictions was below 150 ms, and the models were therefore well-suited for high-frequency trading purposes. Incorporating macroeconomic indicators and sentiment analysis enhanced forecasting accuracy by 1 0-2 0%, establishing the importance of extraneous data sources in predictive modeling. An early stopping criterion at epoch 25 ensured the best training for LSTM to avoid overfitting and improve generalization. Comparative evaluation with conventional ARIMA and Transformer-based models proved that deep learning performs better than statistical forecasting methods in commodity price forecasting. The findings validate that LSTM yields better forecasting accuracy, whereas SVM and RF provide complementary strengths in stability and volatility identification. Future studies need to investigate hybrid deep learning frameworks and Explainable AI (XAI) methods to boost interpretability and continue optimizing commodity price forecasting models. © 2025 IEEE.

A coupon is a ticket or document used in marketing that may be redeemed for a monetary discount or refund when purchasing a product. The problem, in this case, is to know if a customer will accept a coupon for a particular venue. The answers that the user will drive there ‘right away’ or ‘later before the coupon expires’ are labeled as ‘Y = 1’, and the answers ‘no, I do not want the coupon’ is labeled as ‘Y = 0’. This paper proposes integrating three machine learning techniques to create an ensemble boosting classification (EBC) model for a vehicle coupon recommendation. The algorithms used are k-nearest Neighbor (KNN), Support Vector Machine (SVM), and Decision Tree (DT). The results show that the EBC model surpasses the three machine learning models and achieves the highest performance of accuracy 97.37%, precision 94.14%, recall 96.41%, and F1-score 95.28%. © 2024, Penerbit UTHM. All rights reserved.