International Journal of Advances in Soft Computing and its Applications

A Hybrid Deep Neural Network and Morphological Knowledge to Enhance Arabic Lemmatization

Samir BELAYACHI, Azzeddine MAZROUI — Mon, 16 Feb 2026 00:00:00 +0000

Lemmatization is performed in many natural language processing applications during the preprocessing stage, enabling more efficient text analysis and the extraction of relevant information. The Arabic language is associated with the following challenges of lemmatization: the morphological richness of the language, the high usage of concatenations and the omission of the diacritical marks. To overcome those issues, we will introduce a new lemmatizer to improve the functionality of a deep learning framework. The latter will be a BLSTM network, with an additional filtering layer based on morphological features. To alleviate the effect of out-of-vocabulary words, a statistical layer built on Hidden Markov Models was introduced. Tests done on a reference corpus showed that accuracy was enhanced by a margin of over nine percentage points on the use of the two layers (filtering and statistical). In addition, the results were compared with state-of-the-art lemmatizers on an independent corpus which proved that the proposed approach was superior.

Predicting Bitcoin Prices Using Deep Learning

Wed, 11 Feb 2026 00:00:00 +0000

Predicting cryptocurrency price is challenging owing to high volatility, less historical data, and the impact of external parameters like news, public sentiment, and regulatory announcements. This challenge is tackled in this research by employing models of deep learning like Recurrent Neural Network (RNN), Bidirectional Long Short-Term Memory (BiLSTM) and Gated Recurrent Unit (GRU)—to predict Bitcoin's OHLC prices daily. Based on historical time-series data of Coin Codex, the research uses an autoencoder-based feature extraction method with five-day sliding window method for sequence generation. Hyperband optimization is used to tune hyperparameter of each model. The result shows that BiLSTM performs better than all the other models with minimum Mean Squared Error (MSE = 0.001183), Mean Absolute Error (MAE = 0.026090), and maximum R² score (0.980596) after optimization. The results emphasize the significance of deep learning in capturing nonlinear dynamics in time series of financial applications and bear testimony to the effectiveness of hyperparameter tuning in enhancing model accuracy. The study enhances the development of prediction tools for digital asset markets and enables more informed investment decisions.

A Deep Learning Approach for Predicting Hospital Length of Stay for People with Diabetes using Electronic Health Records (EHRs)

Rafat Hammad, Abeer AL-Slamat — Tue, 10 Feb 2026 00:00:00 +0000

Diabetes, which substantially affects the people lives as well as the healthcare systems worldwide, has become the primary area of concern. The purpose of this study is to predict the length of hospital stay among diabetic patients using machine learning (ML), deep learning (DL), along with electronic health record (EHR) data. This work evaluated several models: XGBoost, Random Forest, Multilayer Perceptron (MLP), Convolutional Neural Network (CNN), Recurrent Neural Network with LSTM (RNN-LSTM), and an ensemble model combining GAT and CNN in terms of performance metrics used, such as MAE, MSE, and R². Among all the models tested, MLP using the top 20 features performed the best, recording an MAE of 0.37 and an R² of 0.62. The model emphasized certain critical aspects concerning feature selection and dimensionality reduction, which enhanced the algorithm's accuracy, especially in cases with numerous redundant or verbose variables. More generally, these findings point to the potential that predictive modeling and artificial intelligence have in optimizing the allocation of hospital resources, reducing the costs of health care, and improving patient outcomes, thereby paving the way forward in the development of some clinical decision support tools to manage inpatient diabetic care.

Towards Personalized Lipid Management: Predicting Statins Therapy Eligibility through Machine Learning Models

Tue, 10 Feb 2026 00:00:00 +0000

Cardiovascular diseases are among the leading causes of death worldwide and a major contributor to the deterioration of quality of life. Therefore, it is highly beneficial to follow the clinical guidelines and recommendations for preventing and treating cardiovascular diseases at their early stages. Cholesterol-lowering drugs such as Statins are considered first-line medications for the prevention of atherosclerotic cardiovascular diseases (ASCVD). However, it is not easy to determine patients’ eligibility for statin therapy. In this work, we built efficient and accurate prediction models based on several machine learning algorithms for predicting patients' eligibility for Statins using several cardiovascular disease risk factors. The results indicated that the gradient boosting classifier achieved 95.6% accuracy and 99.0% area under the curve in predicting patients' eligibility for statin therapy. Other simpler but more explainable algorithms such as decision tree and logistic regression also demonstrated good performance.

AI-Powered Security-Aware Reconfiguration in Cyber-Physical Systems for Smart Healthcare and Energy Domains

Theyazn H.H Aldhyani, Rajit Nair, Hasan Alkahtani, Osamah Ibrahim Khalaf — Sun, 08 Feb 2026 00:00:00 +0000

Cyber-physical systems (CPS) in smart energy and smart healthcare must continue to be safe, fast, and reliable, even in the face of evolving cyber threats and fluctuating network/edge conditions. An AI-based, security-aware framework of CPS reconfiguration is presented in this paper. It is comprised of (i) privacy-preserving federated anomaly detection at distributed edge nodes, (ii) a reinforcement-learning decision module that selects risk-aware reconfiguration actions under the objectives of latency, energy, and safety, and (iii) auditability, backed by blockchain, to provide trustworthy governance and accountability in the aftermath of the event. The detection module learns behavioral baselines at the local level and shares only protected updates to support data minimization and privacy compliance while still achieving high accuracy in heterogeneous deployments. The RL controller dynamically modifies actions to reduce service disruption, quicken recovery, and avert unsafe control transitions. A lightweight operational ledger captures reconfiguration activities and trust updates to provide auditable governance in multi-stakeholder CPS settings. In energy and healthcare CPS scenarios, there is strong operational performance. The system achieves 97.8% detection accuracy and an F1-score of 97.4% with an end-to-end latency of 41ms, coupled with a reconfiguration time of 65ms and a mean time to recovery of 5.6 seconds. The framework provides 99.4% uptime, consumes approximately 10 W at the edge, and with low error rates (2.1% false positive and 1.0% false negative), achieves 19.5 Mbps secure throughput. These results demonstrate that self-reconfigurable CPS can maintain mission-critical operational continuity while enhancing privacy, scalability, and governance in large-scale deployments.

Emotion-Aware Adaptive User Interfaces for Enhanced User Experience Using Multi-Modal Deep Learning

Ahmed Alshehri — Thu, 05 Feb 2026 00:00:00 +0000

Emotion-conscious computing is decisive in the further development of human-centered digital interaction, but the direct role in improving the User Experience (UX) has not been studied in detail. This paper introduces an Emotion-Aware Adaptive User Interface (EAAUI) system, which uses multi-modal deep learning to enhance usability, engagement and cognitive efficiency by adapting to emotions in real-time. The suggested solution combines facial expressions, speech prosody, and physiological cues with the help of the hybrid deep learning structure that consists of Convolutional Neural Networks (CNNs), Bidirectional Long Short-Term Memory networks (BiLSTMs), and Transformer-based attention systems to provide strong emotion detection and fusion. Emotions sensed dynamically are the motivators of adaptive interface changes, such as simplifying layout, modulating colors, and providing personal feedback. The framework is tested on benchmark datasets (AffectNet, RAVDESS and DEAP) and controlled user study with 30 people. The experimental findings show that the emotion recognition performance is good with an accuracy of 89.6% and an F1-score of 0.88. According to the UX view, the adaptive interface made task completion (21% less) and error reduction (18% less) significantly faster, user engagement (26% more) higher, and the System Usability Scale (SUS) score was 82.5. The results validate the claim that the operational effect of emotion-aware adaptive interfaces is beneficial to user experience, as they provide viable implications to UX-based applications in the educational, healthcare, and intelligent interactive systems.

A Privacy-Preserving Blockchain and Machine Learning Framework for Secure Next-Generation Genomic Data Analysis

Thu, 05 Feb 2026 00:00:00 +0000

Next-generation genomic data analysis faces ongoing challenges in collaboration, privacy, and scalability. With data protection and no access control deficiency, centralized systems lack sufficient protection for sensitive genomic data. The first-of-its-kind analysis of genomics combined with integrated machine learning and homomorphic encryption with the new privacy-preserving computational framework will be revolutionary. The use of smart contracts, which is unlike other frameworks, for access control, tokenized encrypted, and suspended federated model training during the suspension of other research nodes will be unprecedented. Simulation of genomic datasets (0 Major issues were identified in file sync performance and data protection and security) Vis-a-vis the other frameworks, the Model Proposed outperformed traditional, federated and HE based frameworks with significant. Of the models proposed, this one is the most impressive with a score of 94% precision, 92% recall, 0.93 computed F1 score, and 0.96 Area Under Curve (AUC). The model with the best performance. With 5K genomic datasets in a pilot simulation, collaboration improved by 25% with no breaches in the datasets. The promise of this design to ethically and securely address privacy-preserving genomic data analysis and the subsequent use of Artificial Intelligence in biomedical systems will be groundbreaking.

Comparative Study of LSTM, GRU, and BRNN Performance: large-Scale Data Analytics (EMG Signal Classification)

Thu, 05 Feb 2026 00:00:00 +0000

Electromyography (EMG) signals play a pivotal role in biomedical applications, such as prosthetic control and human-computer interaction, where advanced classification methods for accurate muscle activity translation are essential. This study evaluates the performance of three neural network architectures: Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Bidirectional Recurrent Neural Network (BRNN) for EMG signal classification. The EMG signals were preprocessed using Digital Wavelet Transform (DWT) with Daubechies 2 wavelet to extract time-frequency features. Experiments were conducted on a large-scale training dataset comprising 672 subject recordings across six hand gestures, enabling a robust, data-driven comparison. The best classification accuracy of LSTM, GRU, and Bidirectional RNN was achieved, corresponding to cD7, with values of 93.04±1.52, 92.72±1.26, and 91.59±0.97, respectively. However, these models exhibited varying degrees of sensitivity to additive noise, particularly at deeper DWT levels. The findings highlight the trade-offs between accuracy and noise tolerance, providing insights for optimizing EMG-based gesture recognition systems in real-world applications. The final analysis confirms that LSTM outperforms the other models for real-time EMG classification, while GRU and BRNN offer a favourable balance between accuracy and computational efficiency. Looking ahead, the effective handling of large-scale, high-dimensional EMG data yields significant improvements in performance, particularly for prosthetic and real-time control applications.

Super learner ensemble-based internal quality assessment of watermelon via integration of tapping acoustics and rind texture analysis

Ketsarin Chawgien, Supaporn Kiattisin — Thu, 22 Jan 2026 00:00:00 +0000

Watermelon (Citrullus lanatus) is a widely cultivated fruit recognized for its high sugar content. Accurate detection of maturity and soluble solid content (SSC) is essential to ensure optimal harvest timing, sweetness, and market value, as well as to manage resource usage efficiently. This study introduces a low-cost, portable, and non-destructive approach for maturity classification and SSC estimation in Kinnaree watermelon by integrating tapping acoustics and rind texture analysis with ensemble learning algorithms. Tapping-induced acoustic signals were analyzed to extract key resonant features, while rind texture was quantified using image processing techniques. Selected features from both data sources, combined with watermelon mass, were utilized for three-class maturity classification and SSC regression modeling. Machine learning (ML) algorithms were used to map complex and nonlinear relationships between features and watermelon quality attributes. Results demonstrated that acoustic features and fruit mass were critical for maturity classification. Visual features were essential for SSC estimation. Super learner ensemble demonstrates superior predictive accuracy compared to other models, both in classifying ripeness and predicting the SSC of watermelons. Comparative studies with earlier methods confirmed the effectiveness and competitiveness of the proposed technology for non-destructive evaluation of watermelon quality.

Enhanced Data Embedding Using Adaptive Neural Networks with Modified Whale Optimization Algorithm

Nameer El Emam, Kefaya Qaddoum — Tue, 20 Jan 2026 00:00:00 +0000

This research paper introduces a new steganography algorithm for embedding large amounts of secret messages within a color image. Five security levels are incorporated to ensure reliable protection. To embed data in a randomized way, the algorithm uses a segmentation technique called New Adaptive Image Segmentation (NAIS). By analyzing the properties of each byte, this method determines the appropriate size of secret data to replace each byte and color in the image. Additionally, the algorithm features a machine-learning component inspired by an Adaptive Neural Network (ANN) combined with a modified version of the Whale Optimization Algorithm (MWOA). The findings show that strong imperceptibility is achieved with the stego-image, even with a large payload, reaching four bits per byte (4-bpb) at specific bytes. When the machine learning model ANN_MWOA is used, the highest PSNR reaches 79.58dB for the Baboon color image with a payload of (16384) bits, while PSNR decreases by 1% when applying ANN_WOA. Moreover, the proposed method outperforms previous approaches by an average of 2%. Additional metrics (MSE, SNR, Euclidean Norm, and others) are used to confirm that the proposed algorithm efficiently embeds hidden data.

Modeling Anomalous Diffusion with the Fractional Brusselator

Maysoon Qousini, Waseem Al-Mashaleh — Thu, 12 Feb 2026 00:00:00 +0000

This paper studies the fractional reaction-diffusion Brusselator model, which incorporates fractional-time derivatives to describe memory effects and anomalous diffusion in pattern formation. A fully discrete numerical scheme is developed using an L1 approximation for the fractional derivative and a finite difference method for spatial discretization. Theoretical analysis proves the uniqueness, asymptotic stability, and convergence of the scheme. Numerical simulations demonstrate the emergence of stationary Turing patterns under appropriate conditions, validating the model’s ability to capture complex spatiotemporal dynamics. The work provides a reliable computational framework for exploring fractional reaction-diffusion systems in two dimensions.

Legal Investigation of Financial Fraud Using Artificial Intelligence

Belal Zaqaibeh — Thu, 12 Feb 2026 00:00:00 +0000

Financial fraud remains a serious global challenge, requiring innovative investigative techniques that seamlessly integrate Artificial Intelligence (AI) with legal systems. Conventional fraud investigation methods, which largely rely on AI-driven tools combined with manual legal analysis, often suffer from limited efficiency, scalability, and adaptability when addressing evolving financial fraud patterns. In this study, a hybrid AI-driven framework is proposed to effectively bridge financial fraud investigation and legal analysis. The proposed model integrates AI-based analytical systems with legal reasoning mechanisms to enhance the accuracy and reliability of fraud investigations. The framework has been validated using real-world financial fraud cases and legal investigation records. The results demonstrate that integrating AI-driven systems with fraud investigation processes significantly improves detection accuracy and accelerates investigation timelines. Moreover, the proposed approach enhances the efficiency of financial fraud case analysis by leveraging AI for legal interpretation and decision support. This study also addresses key challenges associated with the application of explainable artificial intelligence (XAI) in judicial contexts. By narrowing the gap between AI-based fraud detection and legal enforcement, the proposed framework contributes to making fraud investigations more transparent, accountable, and legally defensible. The findings confirm that the approach is effective in strengthening forensic financial investigations, reducing investigative effort, and ensuring compliance with continuously evolving financial regulations. Furthermore, sustainability is emphasized as a critical consideration in modern fraud investigation practices. The adoption of artificial intelligence within sustainable legal frameworks may support long-term regulatory compliance and responsible financial governance.