Abstract Accurate house price prediction is essential for real estate valuation, investment planning, and intelligent property decision-support systems. This study proposes an optimized hybrid deep learning framework that integrates a Gated Recurrent Unit and Multilayer Perceptron model with the Binary Whale Optimization Algorithm for feature selection and Ant Colony Optimization for hyperparameter tuning. The proposed framework was evaluated using a publicly available Kaggle house price regression dataset containing 500 housing records with structural, locational, and amenity-related attributes. The dataset was divided into training, validation, and testing subsets using a 70:20:10 ratio, and leakage-free normalization was applied using only the training data. Experimental results show that the proposed BWOA–ACO–GRU–MLP model outperformed standalone GRU, MLP, CNN, LSTM, and BiLSTM models. It achieved an MSE of 0.0146, MAE of 0.1051, RMSE of 0.1208, MAPE of 0.0112, MedAE of 0.0969, and an R 2 of 99.04%. These results demonstrate that combining feature selection, hyperparameter optimization, and hybrid neural regression improves prediction accuracy and model stability for house price estimation. The proposed framework provides a reliable data-driven approach for smart real estate valuation applications.

Accurate earthquake magnitude estimation is critical for mitigating seismic hazards; however, the inherently nonlinear and dynamic characteristics of geophysical data present persistent modeling challenges. To address this complexity, this study proposes a novel hybrid deep learning framework, CNN-DNN-GRU. This architecture synergizes Convolutional Neural Networks (CNNs) for spatial feature extraction, Deep Neural Networks (DNNs) for hierarchical data representation, and Gated Recurrent Units (GRUs) for capturing long-term temporal dependencies. The framework's efficacy is further enhanced by integrating Binary Differential Evolution (BDE) for robust feature selection and the Firefly Algorithm (FA) for rigorous hyperparameter tuning. The model was trained and evaluated on a comprehensive tabular dataset comprising 3,500,000 seismic event instances-detailing parameters such as magnitude, depth, and geospatial coordinates-spanning from 1990 to 2023, utilizing a chronological 70%-15%-15% split for training, validation, and independent testing, respectively. Empirical evaluation demonstrates the proposed model's high predictive accuracy, yielding a Mean Squared Error (MSE) of 0.0109, Root Mean Squared Error (RMSE) of 0.1048, Mean Absolute Error (MAE) of 0.0837, Median Absolute Error (MedAE) of 0.0709, Mean Absolute Percentage Error (MAPE) of 0.0230, and a Coefficient of Determination ([Formula: see text]) of 0.9918. These findings validate the architecture's superior generalization capabilities, establishing it as a highly reliable computational tool for a magnitude-estimation component that may support future early-warning-oriented systems when integrated with real-time waveform and station-level data.

Phishing attacks, which trick users and obtain private data, are still a constant threat to cybersecurity. Using a large dataset of 10,000 samples, each with handcrafted features that capture URL structure and content-based cues, this study suggests an optimized deep learning framework for phishing website detection. The Binary Genetic Algorithm (BGA) is used for feature selection to improve model performance, and Bayesian Optimization (BO) is used for hyperparameter tuning. After normalization, the dataset is separated into subsets for training (70

Sarcasm detection remains a challenging task in natural language processing due to the complex interaction between linguistic and pragmatic cues. While transformer-based models have shown encouraging performance, it is not yet clear whether improvements stem primarily from architectural design or explicit feature engineering. This work introduces a systematic ablation study of three multilingual transformer architectures (mBERT, mDeBERTa-v3, and XLM-RoBERTa) evaluated on HeteroSarc-47K, a curated multi-domain dataset containing 47,694 instances across six heterogeneous domains, including social media and dialectal Arabic text. We isolate the empirical impact of two feature augmentation techniques: emoji integration and contrastive pre-training. Our experiments reveal that while contrastive alignment increases sarcasm recall for mBERT, it induces a significant drop in its overall accuracy, demonstrating architecture-specific sensitivities. In contrast, architectural upgrades alone consistently yield robust improvements; mDeBERTa-v3 achieves the highest performance (72.79% sarcasm recall and 87.20% accuracy) without requiring any feature augmentations, outperforming the baseline mBERT. Furthermore, per-class analysis reveals a persistent performance gap between non-sarcastic and sarcastic classes across all domains, highlighting that class imbalance remains a critical challenge independent of architecture. Ultimately, these findings provide a vital practical takeaway for practitioners: upgrading to inherently stronger multilingual architectures yields significantly higher and more robust performance returns compared to investing in extensive feature-engineering pipelines.

This paper introduces a novel classification model for lung cancer prediction through preprocessing, feature selection (FS) algorithms, and deep learning (DL) models. The public shared dataset from the Kaggle repository was utilized, which consists of 15 attributes associated with lung cancer. Data preprocessing tasks included addressing issues such as missing values, label encoding, and standardization. SMOTE was used to address class imbalance. Five FS algorithms in binary version: Particle Swarm Optimization (BPSO), Genetic Algorithm, Grey Wolf Optimizer, Whale Optimization Algorithm, and Ant Colony Optimization were utilized to evaluate performance. The BPSO achieved the best accuracy with 98.15

Accurate vehicle-level CO₂ emission prediction is essential for regulatory planning and greener powertrain design. This study presents an integrated FA-ACO-LSTM framework for estimating vehicle-level tailpipe CO₂ emissions from standard vehicle descriptors. The contribution lies in combining leakage-aware Binary Firefly feature selection, mixed-variable FA-ACO hyperparameter tuning, repeated-run evaluation, and post hoc interpretability into a unified prediction pipeline. Using a public fuel-consumption dataset sourced from the official Canadian vehicle fuel-consumption and emissions registry (open government portal) (22,556 vehicles), we adopt a 70/20/10 train/validation/test split and perform wrapper-based feature selection with Binary Firefly Algorithm (BFA). The FA-ACO scheduler jointly tunes hyperparameters for five candidate learners (LSTM, CNN, GRU, MLP, TabNet) and selects the best. FA-ACO-LSTM attains MSE = 0.0099, MAE = 0.0791, MedAE = 0.0664, MAPE = 0.84%, and R² = 98.71% on the test set, surpassing FA-ACO-CNN (R² = 93.55%), FA-ACO-GRU (90.03%), FA-ACO-MLP (87.60%), and FA-ACO-TabNet (84.88%). Compared with untuned baselines, FA-ACO improves LSTM from R² = 96.53% to 98.71% and reduces MSE by 45%. BFA yields the lowest average error among feature selectors versus BMWO, BFO, and BGWO, and an ANOVA across models confirms significance (F = 30.2, p < 0.0001). These results represent a fuel-consumption-informed prediction scenario because the selected feature set includes COMB (L/100 km), HWY (L/100 km), and COMB (mpg), which are strongly associated with certified CO₂ emissions. When all fuel-consumption indicators were excluded before feature selection and training, model performance decreased from [Formula: see text]to [Formula: see text]. The ablation confirms that fuel-consumption proxies account for a substantial part of the main predictive performance, while non-consumption vehicle attributes retain meaningful predictive value, and indicates that, under the adopted fixed feature-ordering representation and FA-ACO tuning protocol, LSTM achieved the best performance among the evaluated neural and tabular baselines. However, because the dataset is cross-sectional rather than temporal, the LSTM should be interpreted as a feature-interaction learner rather than a temporal forecasting model.

Breast cancer (BC) ranks among the significant causes of death rates in women globally; therefore, early and precise diagnosis is essential for efficient treatment and survival. Machine learning (ML) tools have improved BC diagnosis. This paper introduces an enhanced BC classification model that uses hybrid PSO–GWO optimization after stacking ML models. Using a dataset of BC shared online at the Kaggle repository, which consists of 4024 patients and 16 attributes, the paper applies pre-processing tasks to prepare the BC dataset. For feature selection (FS), binary versions of Particle Swarm Optimization (BPSO), Grey Wolf Optimizer (BGWO), and Whale Optimization Algorithm (BWAO) were utilized. The BPSO obtained the best fit to the dataset with 0.92025. This study employs six ML models, such as Random Forest (RF) Classifier, Stochastic Gradient Descent (SGD) Classifier, Support Vector Machine (SVM), Gradient Boosting Classifier (GBC), k-Nearest Neighbors (k-NN), XGBoost Classifier (XGB), and Naïve Bayes Classifier (NBC) to evaluate the proposed approach. The RF achieved the best accuracy of 89.22

Diabetes is a chronic condition that affects a substantial portion of the global population and is linked to elevated mortality rates and a range of severe health complications. Despite its clinical importance, progress in diabetes research is often constrained by the limited availability of comprehensive datasets and robust predictive models. To address these challenges, researchers are increasingly turning to big data analytics and machine learning (ML) methodologies. This study presents the development of an ML-based system aimed at predicting the likelihood of diabetes and classifying its various types. A novel dataset, termed Diabetes Types Dataset, was constructed by integrating four heterogeneous dataset sources: paediatrics data from the Mansoura University Children Hospital repository, the Pima Indian Diabetes (PIMA) dataset, the Pone dataset, and a Gestational Diabetes dataset. The classification of diabetes types was approached as a multiclass problem using a suite of supervised ML algorithms, including Artificial Neural Networks (ANN), Logistic Regression, Naive Bayes, Decision Trees, Adaptive Boosting, Random Forests, Gradient Boosting, Support Vector Machines, and K-Nearest Neighbors. Model performance was evaluated using several metrics: Accuracy, Precision, Mean Squared Error, and Area Under the Receiver Operating Characteristic Curve. Among the models tested, the ANN classifier demonstrated the highest accuracy, achieving a peak performance of 99.98%. Further validation was conducted using an external dataset referred to as diabetes_prediction, which confirmed the model’s robustness with consistent accuracy. Additionally, the proposed system was applied to a publicly available dataset, diabetes_Dataset, containing 34 features used to predict 12 distinct types of diabetes efficiently. The results suggest that this ML-driven approach can significantly enhance the ability of healthcare professionals to detect and classify diabetes types, thereby supporting early intervention and improved disease management.

Air quality prediction is an important environmental task that supports pollution monitoring, public health protection, and sustainable urban decision-making. This study proposes a hybrid BiGRU-MLP prediction framework enhanced using Binary Ant Colony Optimization (BACO) for feature selection and the Firefly Algorithm (FA) for hyperparameter tuning. The proposed model was evaluated using the public Air Quality dataset from Kaggle, which contains 9,357 records and multiple pollutant and meteorological attributes, including CO(GT), NOx(GT), NO2(GT), temperature, relative humidity, and sensor-based air quality indicators. The dataset was divided into 70% training, 20% validation, and 10% testing subsets to ensure reliable model development and unbiased performance evaluation. Several baseline models were implemented for comparison, including BiGRU, MLP, Neural Network (NN), LSTM, and BiLSTM. The experimental results demonstrate that the proposed BiGRU-MLP model achieved the best predictive performance, obtaining an MSE of 0.0001, RMSE of 0.0101, MAE of 0.0081, MedAE of 0.0070, MAPE of 0.0008, and R² of 99.99%. In comparison, the standalone BiGRU achieved an R² of 96.87%, while MLP, NN, LSTM, and BiLSTM achieved R² values of 95.17%, 92.71%, 90.88%, and 89.29%, respectively. These results confirm that combining BiGRU’s temporal learning capability with MLP’s nonlinear feature representation significantly improves prediction accuracy. The integration of BACO and FA further enhances the model by selecting informative features and optimizing model parameters. Therefore, the proposed framework provides an accurate and efficient solution for air quality prediction and can support intelligent environmental monitoring systems.

Potable water is essential for human and environmental health. Traditional water quality assessments are slow and uncertain. To improve water potability classification, we propose a hybrid deep learning model. The architecture combines Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM) networks for enhanced temporal feature learning. In this paper, we utilized the Water Potability dataset from Kaggle and applied hot deck imputation to handle missing values. To address class imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) was applied to the training data, enhancing the model’s ability to learn from underrepresented potable water samples. To improve model performance, the Relief Algorithm selected features and Bayesian Optimization (BO) to optimize the hyperparameters of GRU-LSTM. The dataset was 70

The widespread use of credit cards has led to an increase in fraud. Credit card fraud detection involves identifying and preventing fraudulent transactions, either in real-time or post-occurrence. This paper seeks to create an advanced credit card fraud detection model via data mining. The proposed method comprises four essential steps: data acquisition, preprocessing, feature selection, and fraud detection. A recent balanced dataset is acquired, containing 28 anonymized features about the credit card transactions, along with the transaction amount and the transaction label (normal or fraud). The dataset is then explored to clean it and ensure its integrity. Feature selection is executed via the Energy Valley Optimization (EVO) metaheuristic method, employing the accuracy value of the Light Gradient Boosting Machine (LGBM) as the fitness function. This results in a 30% reduction in features. The reduced dataset is then input into the classification step, where an ensemble soft voting model is applied. This model encompasses Extra Trees, eXtreme Gradient Boosting (XGBoost), and Categorical Boosting (CatBoost) classifiers. The proposed model averages the probability of the three classifiers for each label and outputs the label with the highest average probability. The proposed method is assessed using recall, precision, accuracy, and F1-score, attaining 99.89%, 99.58%, 99.74%, and 99.74%, respectively. The proposed approach is evaluated against existing machine learning classifiers and relevant studies using the same dataset, showcasing enhanced performance and confirming its efficacy in identifying credit card fraud.

BACKGROUND:Gelotophobia, the fear of being laughed at, is a social anxiety condition that affects approximately 6% of neurotypical individuals and up to 45% of those with autism spectrum disorder (ASD). This comorbidity can significantly impair the quality of life, particularly in adolescents with high-functioning ASD, where the prevalence reaches 41.98%. Accurate and automated detection tools could enhance early diagnosis and intervention. OBJECTIVE:This study aimed to develop a deep learning-based diagnostic system that integrates facial emotion recognition with validated questionnaires to detect gelotophobia in individuals with or without ASD. METHODS:The system was trained to identify ASD status using a balanced dataset of 2932 facial images (n=1466; 50% from individuals with ASD and n=1466; 50% from neurotypical individuals). The images were processed using the DeepFace library to extract facial features, which were then used as input for the deep learning classifier. After identifying ASD status, the same images were further analyzed using the pretrained DeepFace model to evaluate facial expressions for signs of gelotophobia. In cases where facial cues were ambiguous, the GELOPH<15> questionnaire, consisting of 15 items, was administered to confirm the diagnosis The system was fully implemented using the Python programming language. Deep learning models were developed using libraries such as PyTorch for training the multilayer perceptron classifier, while CUDA was used to accelerate computations on compatible graphics processing units. Additional libraries from the Python programming language, such as scikit-learn, NumPy, and Pandas, were used for preprocessing, model evaluation, and data manipulation. DeepFace was integrated using its Python application programming interface for facial recognition and emotion classification. RESULTS:The dataset comprised 2932 facial images collected from platforms such as Kaggle and ASD-related websites, including 1466 (50%) images of children with ASD and 1466 (50%) images of neurotypical children. The dataset was split into 2653 (90.48%) training samples and 279 (9.51%) testing samples, with each image contributing 100,352 extracted features. We applied various machine learning models for ASD identification. The system achieved an overall prediction accuracy of 92% across both training and testing datasets, with the multilayer perceptron model demonstrating the highest testing accuracy. The system successfully classified gelotophobia in cases where facial expressions were clear. However, in cases of ambiguous facial cues, the DeepFace model alone was insufficient. Incorporating the GELOPH<15> questionnaire improved diagnostic reliability and consistency. CONCLUSIONS:This study demonstrates the effectiveness of combining deep learning techniques with validated diagnostic tools for detecting gelotophobia, particularly in individuals with ASD. The high accuracy achieved highlights the system's potential for clinical and research applications, contributing to the improved understanding and management of gelotophobia among groups considered socially vulnerable. Future research could expand the system's applications to broader psychological assessments.

In recent years, the continuous increase in the growth of text data on social media has been a major reason to rely on the pre-training method to develop new text classification models specially transformer-based models that have proven worthwhile in most natural language processing tasks. This paper introduces a new Position-Context Additive transformer-based model (PCA model) that consists of two-phases to increase the accuracy of text classification tasks on social media. Phase I aims to develop a new way to extract text characteristics by paying attention to the position and context of each word in the input layer. This is done by integrating the improved word embedding method (the position) with the developed Bi-LSTM network to increase the focus on the connection of each word with the other words around it (the context). As for phase II, it focuses on the development of a transformer-based model based primarily on improving the additive attention mechanism. The PCA model has been tested for the implementation of the classification of health-related social media texts in 6 data sets. Results showed that performance accuracy was improved by an increase in F1-Score between 0.2 and 10.2% in five datasets compared to the best published results. On the other hand, the performance of PCA model was compared with three transformer-based models that proved high accuracy in classifying texts, and experiments also showed that PCA model overcame the other models in 4 datasets to achieve an improvement in F1-score between 0.1 and 2.1%. The results also led us to conclude a direct correlation between the volume of training data and the accuracy of performance as the increase in the volume of training data positively affects F1-Score improvement.

Enhancing hydrogen-based energy systems requires accurate hydrogen consumption forecast. This paper compares random forest regressor, multi-layer perceptron, support vector regressor, and gradient boosting regressor to forecast hydrogen consumption, using production and consumption capacity and geographical coordinates such as production capacity, consumption capacity, latitude, and longitude from the European Hydrogen Observatory, was analyzed using advanced statistical techniques to ensure robust preprocessing and feature selection. It encompasses hydrogen consumption data from various production pathways (e.g., electrolysis and steam reforming), as reported by the European Clean Hydrogen Observatory, although specific production methods are not individually labeled. The study evaluates model performance using standard regression metrics, including mean absolute error, mean squared error, root mean squared error, coefficient of determination, and median absolute error. The random forest regressor led with 0.9789 and low error metrics (mean absolute error = 0.0010, mean squared error = 0.0030, and root mean squared error = 0.0034). However, to further improve prediction accuracy, a stacking ensemble model was developed by combining random forest regressor, multi-layer perceptron, support vector regressor, and gradient boosting regressor as base learners, with Ridge regression serving as the meta-learner. The stacked model significantly outperformed all individual models, achieving a coefficient of determination score of 0.9963, with a reduction in error metrics (mean absolute error = 0.0009, mean squared error = 0.0002, and root mean squared error = 0.0014). To gain deeper insights into feature importance, SHapley Additive Explanations analysis was conducted on the stacked model. The results indicate that latitude, longitude, and production capacity are the most influential factors affecting hydrogen consumption. This paper indicates that the stacked model can effectively anticipate hydrogen usage, helping researchers and policymakers optimize hydrogen distribution and consumption strategies for sustainable energy planning.

Clean water requires accurate water quality categorization. A water potability (WP) dataset with pH, hardness, solids, chloramines, sulfate, conductivity, and other metrics for 3276 water bodies was used in this paper. After median imputation for missing values, normalization for feature scaling, and class imbalance correction using SMOTE, the Kaggle public dataset was prepared. With binary particle swarm optimization (BPSO) and binary whale optimization algorithm (BWAO), feature selection (FS) was used to determine the most important features for classification. A subset of seven essential characteristics is selected with the lowest average error of 0.3745 by the BPSO. Random forest (RF), gradient boosting (GB), support vector machine (SVM), Extra Tree (ET), decision tree (DT), and XGBoost are tested for WP prediction. The ET classifier ranked first, with 70.63

This study proposed a novel technique for early diabetes prediction with high accuracy. Recently, Deep Learning (DL) has been proven to be expeditious in the diagnosis of diabetes. The supported model is constructed by implementing ten hidden layers and a multitude of epochs using the Deep Neural Network (DNN)-based multi-layer perceptron (MLP) algorithm. We proceeded to meticulously fine-tune the hyperparameters within the fully automated DL architecture to optimize data preprocessing, prediction, and classification using a novel dataset of Mansoura University Children's Hospital Diabetes (MUCHD), which allowed for a comprehensive evaluation of the system’s performance. The system was validated and tested using a sample of 548 patients, each with 18 significant features. Various validation metrics were employed to ensure the reliability of the results using cross-validation approaches with various statistical measures of accuracy, F-score, precision, sensitivity, specificity, and Dice similarity coefficient. The high performance of the proposed system can help clinicians accurately diagnose diabetes, with a remarkable accuracy rate of 99.8%. According to our analysis, implementing this method results in a noteworthy increase of 0.39% in the overall system performance compared to the current state-of-the-art methods. Therefore, we recommend using this method to predict diabetes.

Diagnosing depression and anxiety involves various methods, including referenda-based approaches that may lack accuracy. However, machine learning has emerged as a promising approach to address these limitations and improve diagnostic accuracy. In this scientific paper, we present a study that utilizes a digital dataset to apply machine learning techniques for diagnosing psychological disorders. The study employs numerical, striatum, and mathematical analytic methodologies to extract dataset features. The Recursive Feature Elimination (RFE) algorithm is used for feature selection, and several classification algorithms, including SVM, decision tree, random forest, logistic regression, and XGBoost, are evaluated to identify the most effective technique for the proposed methodology. The dataset consists of 783 samples from patients with depression and anxiety, which are used to test the proposed strategies. The classification results are evaluated using performance metrics such as accuracy (AC), precision (PR), recall (RE), and F1-score (F1). The objective of this study is to identify the best algorithm based on these metrics, aiming to achieve optimal classification of depression and anxiety disorders. The results obtained will be further enhanced by modifying the dataset and exploring additional machine learning algorithms. This research significantly contributes to the field of mental health diagnosis by leveraging machine learning techniques to enhance the accuracy and effectiveness of diagnosing depression and anxiety disorders.

In today’s era of Big Data, with the integration of data from various systems, devices, and machines used by healthcare service providers, health insurance companies, and their sub-sectors, maintaining privacy and security has become crucial. It is important to uphold the confidentiality and security of data exchanged between data service providers and insurance companies as required by law. The purpose of this paper is to focus on addressing the security and privacy issues associated with healthcare data, particularly concerning medical data in both in- transit and at-rest modes. We aim to provide a proposed solution to enhance data security and maximize privacy protection.

Technical debt (TD) is a metaphor for technical problems that are hidden from customers and end users but hinder development efforts during system evolution and future improvements. TD is frequently incurred due to tight budgets and timelines, which lead to financial, quality, and technical issues that make future maintenance more costly or impossible. Project management concentrates on external issues related to customer satisfaction and does not concentrate on system internal software quality defects and maintenance that be subject to future interest payments. In this paper, we highlight and investigate, in detail, an overview of the Technical Debt items for the software development team in SMEs. In addition, we present a TD management (TDM) approach with best practices for each step, provided with studied cases for applying the TDM approach at software industrial companies. The study shows that the proposed approach to measuring the impact of internal / external software quality leads to increased awareness of TD’s occurrence and provides processes to identify, prevent, monitor, prioritize, and repay TD to satisfy both software technical requirements and customer value and halt project failures and overruns. The analysis of applying our TDM approach leads to a better comprehension of TD contraction in selected software companies, improvements in team morale and motivation, and an increase in its maintainability.

The novel coronavirus disease (COVID-19), which appeared in Wuhan, China, is spreading rapidly worldwide. Health systems in many countries have collapsed as a result of this pandemic, and hundreds of thousands of people have died due to acute respiratory distress syndrome caused by this virus. As a result, diagnosing COVID-19 in the early stages of infection is critical in the fight against the disease because it saves the patient's life and prevents the disease from spreading. In this study, we proposed a novel approach based on transfer learning and deep support vector data description (DSVDD) to distinguish among COVID-19, non-COVID-19 pneumonia, and intact CT images. Our approach consists of three models, each of which can classify one specific category as normal and the other as anomalous. To our knowledge, this is the first study to use the one-class DSVDD and transfer learning to diagnose lung disease. For the proposed approach, we used two scenarios: one with pretrained VGG16 and one with ResNet50. The proposed models were trained using data gathered with the assistance of an expert radiologist from three internet-accessible sources in end-to-end fusion using three split data ratios. Based on training with 70%, 50%, and 30% of the data, the proposed VGG16 models achieved (0.8281, 0.9170, and 0.9294) for the F1 score, while the proposed ResNet50 models achieved (0.9109, 0.9188, and 0.9333).