This study presents a novel optimal energy management strategy for interconnected microgrids under renewable energy uncertainty, leveraging a Whale Optimization Algorithm (WOA)-based scheduling framework. The proposed method is distinctive in that it combines a local power exchange system with mixed-integer linear programming (MILP) to allow the sharing of energy in a cooperative manner while DERs do not have to bear any extra operational costs. Unlike conventional single-microgrid scheduling models, the developed framework explicitly models stochastic solar and wind generation using Weibull and Normal probabilistic distributions and incorporates reliability metrics Loss of Power Supply Probability (LPSP) and Loss of Load Expectation (LOLE) into the objective function. To reduce the uncertainty of demand predictions within 24 hours, a dynamic neural network-based load forecasting method is also used. The simulation outcomes for five microgrids situated at different locations show that the cooperative local system extensively raises the profit level, lowers the total operational costs, and enhances the system reliability by the elimination of the shortage of loads to zero.A comparative study of three different load scenarios has revealed that the combination of WOA and local exchange results in better cost savings and higher energy-sharing benefits than non-cooperative configurations. The new approach is a scalable and resilient tool for on-demand multi-microgrid energy management under renewable energy fluctuations, thus contributing to the economic and reliability aspects of smart grid systems.
This study conducts a systematic literature review (SLR) to assess the state-of-the-art in Software Effort Estimation (SEE), a capability that underpins effective software project management. Reliable effort estimates have a direct influence on budget allocation, resource planning, and delivery schedules, making progress in SEE both practically and scientifically significant. The review’s primary objective is to identify current trends, surface emerging challenges, and delineate opportunities for advancing research in SEE. Following the structured guidelines of Kitchenham (2007), the review applies a transparent protocol for study identification, selection, and synthesis, resulting in a curated set of 87 studies that collectively inform the research questions. Results indicate that CI-based techniques have gained prominence in recent SEE literature, while machine learning and artificial neural networks are the most adopted approaches. The popularity of these methods can be partly explained by their ability to model complex nonlinear relationships among cost drivers and project attributes. However, key challenges persist. Issues related to the availability and quality of datasets are at the top: limited representativeness, inconsistency, and noise limit comparability across studies and reduce external validity. The trade-off between the accuracy of a model and its interpretability remains another pressing challenge faced by practitioners in the field, since highly accurate "black-box" models are often hard to explain and make part of a decision. As stated, the contribution of future studies in assembling more representative datasets and incorporating other AI technologies, including deep learning and natural language processing, is required to further improve the modeling capabilities of SEE. Such improvements are likely to contribute to making estimation methods more robust, scalable, and interpretable, which in turn would advance planning and execution for software projects.
Optimizing concrete mixtures for both compressive strength and cost is critical for achieving economically and structurally efficient designs in the construction industry. Traditional optimization methods, while effective, are limited by tedious procedures and difficulty capturing the complex relationships between mixture components and resulting concrete properties. This study introduces a novel hybrid approach that combines machine learning (ML) with evolutionary optimization to address these limitations. Gradient Boosting and Support Vector Regression (SVR) are employed to model non-linear relationships between input parameters and concrete performance metrics, namely compressive strength and mixture cost. The Human Evolutionary Optimization Algorithm (HEOA) is applied to determine optimal material ratios that satisfy strength requirements while minimizing cost. Sensitivity analysis using ANOVA identifies cement and superplasticizer as the most influential factors, with cement primarily affecting strength and superplasticizer affecting cost. Among the tested models, the hybrid Gradient Boosting–HEOA (GBHE) approach achieved the best performance, with RMSE values of 3.489 for strength and 2.049 for cost, and R² values of 0.964 and 0.997, respectively, outperforming standalone ML models. This work demonstrates a practical framework for designing cost-effective, high-strength concrete mixtures using an integrated ML–optimization strategy, offering significant potential for industrial application.
Accurate short-term building cooling load forecasting is essential for a wide range of building energy management activities, including demand-side management, control optimization, and fault analysis and diagnosis. Traditional forecasting methods often face limitations due to underlying physical assumptions, which can reduce their predictive reliability. In contrast, data-driven techniques utilize large volumes of data and flexible modeling approaches, offering significant improvements in forecast accuracy through advancements in data science and machine learning. This study investigates the application of Lasso Regression (LAS) for predicting building cooling loads, focusing on enhancing its predictive capability through hybrid optimization. Two advanced metaheuristic optimization algorithms, Bald Eagle Search Optimization (BESO) and the Puma Optimizer (PO), were employed to fine-tune the LAS model parameters, resulting in two hybrid frameworks: BES_LASSO, combining LAS with BESO, and PUMA_LASSO, integrating LAS with PO. Experimental evaluation demonstrates that BES_LASSO achieved the highest R² value of 0.8794 in the testing phase, followed by PUMA_LASSO with 0.8425. Similarly, in terms of Root Mean Square Error (RMSE), BES_LASSO outperformed all models with a value of 3.2726, while the standard LAS model exhibited the weakest performance at 4.5352. These results indicate that the integration of advanced optimization techniques substantially improves model accuracy and reliability. The study highlights the potential of hybrid LAS-based models as practical and effective tools for real-time building energy management, enabling more precise control, better energy efficiency, and improved decision-making in both operational and strategic planning contexts.
Recycled Aggregate Concrete (RAC) plays an essential role in sustainable construction by reusing materials from demolished structures and reducing environmental pollution. However, accurately predicting its elastic modulus remains a major challenge due to the variability in material composition and the limitations of traditional empirical formulas. This study aims to improve the prediction accuracy of RAC’s elastic modulus using hybrid machine learning and optimization techniques. Two advanced machine learning models, Adaptive Boosting Regression (ADAR) and Stacking Regression (Stacking), were implemented for their strong learning capabilities. To enhance the predictive performance of these models, the Electrostatic Discharge Algorithm (EDA) was used as an optimizer, creating two hybrid frameworks: ADED (ADAR + EDA) and STED (Stacking + EDA). These models were trained and validated using a comprehensive dataset, and their performance was evaluated through statistical metrics, including R², RMSE, MSE, WAPE, and N10_Index. The results showed that the ADED model achieved the highest accuracy, outperforming other models, while the STED model ranked second with R² = 0.982, RMSE = 0.953, MSE = 0.909, WAPE = 0.040, and N10_Index = 0.955. The findings confirm that hybrid models combining machine learning and optimization significantly improve the prediction of the elastic modulus of RAC, providing a reliable approach for sustainable design and structural analysis in modern construction.
Compressive strength prediction in steel fiber reinforced concrete (SFRC) is challenging in current civil engineering practice due to its highly nonlinear and multi-factor-dependent nature. Modeling it with precise accuracy is important in enabling optimal design mixes and improving the structure's reliability in concrete applications. This study introduces a feature-based regression method with the power of Categorical Boosting Regression (CATR) in collaboration with effective metaheuristic optimizers. Particularly, the Archimedes Optimization Algorithm (AOA) and the Ali Baba and the Forty Thieves Algorithm (AFTO) are used to optimize CATR's hyperparameters and create three hybrids: CAAF (CATR-AFTO), CAAO (CATR-AOA), and CAFO (an ensemble of CAAF and CAAO in a hybrid). These hybrids aim to improve predictive accuracy by merging the power of the base learner with that of the optimizers. The CAFO model is a specially developed ensemble approach that combines the outputs of CAAF and CAAO to leverage their complementary behaviors. This study adopts a feature-based analytical approach and seeks to examine input features' separate and joint impacts on the model performance and determine the most important influencing variables contributing to predictive performance through systematic evaluation of feature importance. The outcomes illustrate that model improvement enhances model performance, optimizes the model's sensitivity to particular features, and provides increased insights into the data structure. In addition, comparing the models shows that CAFO provides better generalization due to its nature of being an ensemble and optimally balanced feature response. These results imply that precise feature inference and optimal hybrid modeling can considerably enhance regression performance and provide more transparent insight into the feature-to-output relationships. This end-to-end machine learning and optimization unification creates novel paths toward constructing model interpretation and high-performance predictive models.
Reinforced concrete (RC) columns are key load-bearing elements in buildings and infrastructures, and their seismic performance is primarily characterized by Drift and maximum shear strength (Vmax). Accurate prediction of these parameters is essential for achieving reliable and cost-effective structural designs. In this study, advanced machine learning (ML) models Random Forest (RFR) and Extreme Gradient Boosting (XGB) are developed and enhanced using two recent metaheuristic optimization algorithms: the Mountain Gazelle Optimizer (MGO) and the Coot Optimization Algorithm (COA). The proposed hybrid frameworks aim to improve prediction accuracy and generalization capability for RC column seismic parameters. Comparative analysis revealed that among standalone models, XGB exhibited superior predictive capability over RFR. Furthermore, the hybrid XGMG model (XGB integrated with MGO) achieved the best overall performance, recording R² values of 0.985 for V max and 0.943 for Drift, along with low RMSE values of 15.81 and 0.161, respectively. These findings confirm that integrating metaheuristic optimizers with ML algorithms significantly enhances model efficiency and stability. The developed hybrid model can serve as a reliable tool for engineers and designers in accurately assessing seismic performance, optimizing material use, and ensuring structural safety. Ultimately, this approach contributes to advancing data-driven methodologies for seismic evaluation of RC columns and can be extended to other structural components.
Diabetes is a metabolic disorder characterized by high blood glucose levels and is a major health burden worldwide due to serious complications such as cardiovascular disease and chronic renal failure. The increasing prevalence of diabetes underscores the importance of early diagnosis and effective treatment. In this context, machine learning (ML) and data mining (DM) techniques can support data-driven diabetes prediction. This study evaluates Histogram Gradient Boosting Classifier (HGBC) and Extreme Gradient Boosting Classifier (XGBC) optimized using Trochoid Search Optimization (TSO) and Sunflower Optimization Algorithm (SOA). Experiments are conducted on the widely used Pima Indians diabetes dataset (768 records, 8 clinical features). Model performance is assessed using accuracy, precision, recall, and F1-score on mutually exclusive training (70%) and testing (30%) subsets to reduce information leakage. Among the evaluated models, the hybrid XGTS model achieved the best performance (accuracy = 0.941), followed by XGSO (accuracy = 0.935), demonstrating that metaheuristic optimization can improve gradient boosting–based classification. SHAP-based interpretation is used to explain the best model and identify influential predictors. These results highlight the potential of hybrid ML frameworks as supportive tools for early diabetes identification, while emphasizing that validation on larger and more diverse populations is required before clinical deployment in practice.
Fake news is becoming a major problem in society today. Fake news is spreading more widely every day, and identifying it is getting harder due to the ease of access to the Internet, the volume and speed of news delivery, and the ever-expanding virtual area. Distinguishing fake and non-fake news is important and practical. Therefore, in this investigation, an attempt was made to present a model for classifying news articles utilizing text analysis techniques and models based on NNs. In this model, textual data is classified into two classes: real and fake. For this purpose, while pre-processing the textual data, the word2index function was used for the Embedding Layer to digitize the textual data. Also, by using a diverse arrangement of different layers in NNs, six different models, including 1CNN, 2CNN, 3CNN, 1CNN-LSTM, 2LSTM, and 1DNN, were formed. The difference between these models is the difference in the composition and dimensions of the layers used in them. The models were evaluated using key performance metrics, including accuracy, recall, precision, and AUC. The proposed 1CNN model had the best index values when compared to other models, according to a case study that compared many assessment indices. Stated differently, the most accurate method for classifying text data is to use a one-dimensional neural network model with a kernel size of two and a convolution layer using sixty-four filters.
Stock price prediction is a critical task in the financial sector due to its profound implications for traders and investors. This paper presents a comparative analysis of machine learning models applied to stock price prediction using historical data from the Nasdaq stock index spanning the years 2015 to 2023. The study introduces an Extreme Gradient Boosting Regression (XGBR) model optimized with three distinct metaheuristic algorithms: Battle Royal Optimization (BRO), Moth Flame Optimization (MFO), and Artificial Bee Colony (ABC). These optimization techniques aim to enhance the model's predictive performance by improving parameter tuning and model generalization. Among the optimized models, the ABC-XGBR demonstrated superior performance due to its strong balance between exploration and exploitation and its effective search capability in high-dimensional feature spaces. The experimental results show significant improvements over the baseline XGBR model, with R² values of 0.9721 for BRO-XGBR, 0.9885 for MFO-XGBR, and 0.9936 for ABC-XGBR. These outcomes underscore the effectiveness of combining machine learning with nature-inspired optimization algorithms to produce more accurate stock price forecasts. This research contributes valuable insights into the practical application of hybrid models for financial forecasting, emphasizing their utility in enhancing predictive accuracy. It also offers decision-makers—such as investors, analysts, and financial institutions—a robust framework for incorporating data-driven strategies into risk assessment and portfolio management. Future work may explore additional datasets, real-time prediction capabilities, and further refinement of optimization algorithms to extend the applicability of these methods to broader financial contexts.
The widespread utilization of social media has precipitated a notable upsurge in the dissemination of inaccurate information. This underscores the urgency to counteract the propagation of falsehoods and decrease the reliance on these platforms as sources of trustworthy information. False content gains traction through user engagement, distorting perceptions of major events and political narratives. In response to this challenge, the field of Natural Language Processing (NLP) and Machine Learning (ML), coupled with Deep Learning (DL) techniques, has arisen to identify and flag fabricated news. However, the task of differentiating authentic news from counterfeit news remains intricate due to the scarcity of reliable datasets and the sheer volume of data present on platforms such as Twitter. In the pursuit of a solution, the development of user-friendly classifiers emerges as pivotal in upholding the integrity of information in the digital era. Therefore, this study delves into the classification of Fake and True news through the employment and comparison of prominent techniques such as the transformer-based GPT-3, Convolutional Neural Networks (CNN), and Long Short-Term Memory (LSTM). The findings reveal that GPT-3 exhibits remarkable parity in terms of accuracy, precision, recall, f1-score, and roc-auc, with values nearly equivalent to one, outperforming other models in this context.
These days, machine learning (ML) is one of the most important technologies, especially in the field of healthcare. Alongside the rapid growth of high-quality medical data and information, its importance is growing. Even with these advancements, early and accurate disease identification is still a difficult task. ML models are trained on historical healthcare data to learn patterns and relationships between different variables. Once trained, these models can make predictions about future events or outcomes, such as the likelihood of a patient developing a particular disease, the risk of complications, or the effectiveness of treatment. Based on the unique traits and medical background of each patient, ML models can forecast the likelihood that a patient would suffer from a certain disease or have a negative health event. For individuals who are more likely to require therapy, this data may be utilized to tailor treatment regimens, assign resources, and prioritize treatment. In this study, ML has been used to support healthcare prediction through the use of two models: Extra Trees Classification (ETC) and Support Vector Classification (SVC). To improve the accuracy of the findings derived from the models, two optimizers Smell Agent Optimization (SAO) and Crocodile Hunting Strategy (CHS) used in combination with two models, SVC, ETC. Four models ETSA, ETCH, SVSA, and SVCH were developed in this study by integrating the ETC and SVC models with two optimization algorithms the CHS and SAO. Upon analyzing the performance of these models, it was found that the ETSA model had a high precision of 0.806 in normal conditions.
One of the most serious medical illnesses impacting a large number of people globally is diabetes mellitus. This condition is influenced by several factors, including advanced age, obesity, sedentary lifestyle, hereditary susceptibility, bad eating habits, and high blood pressure. Diabetes patients are more vulnerable to heart disease, renal failure, stroke, visual impairment, and nerve damage, among other consequences. In modern hospital settings, diagnosing diabetes entails gathering a wealth of relevant data via a series of tests, allowing medical practitioners to customize treatment regimens. In this regard, integrating big data analytics has become an essential tool for the healthcare industry. Healthcare firms use big data analytics to dive into large datasets and find hidden patterns and insights since they have access to enormous amounts of data. This analytical skill enables practitioners to extract meaningful insights from the data, which supports well-informed decision-making and accurate result prediction. Support Vector Classification (SVC) was employed to predict Diabetes in this study. Additionally, 3 novel metaheuristic algorithms, Quadratic Interpolation Optimizer (QIO), Tunicate Swarm Algorithm (TSA), and African Vulture Optimization Algorithm (AVOA) were utilized to enhance the SVC’s performance. The fundamental model was combined with 3 optimization approaches to create the hybrid models: SVC + QIO (SVQI), SVC + AVOA (SVAV), and SVC + TSA (SVTS). When it came to accuracy metric values during testing, the SVQI model performed best with a value of 0.877. The SVAV and SVTS models both secured the second-best performance in this segment with values of 0.871.
Accurate and reliable runoff forecasts are essential for effective water resource management and flood control operations. Hydrological forecasting plays a key role in decision-making, especially under changing climate conditions. Recent advances in machine learning (ML) have opened new opportunities to improve prediction accuracy. This study focuses on evaluating commonly used ML methods for runoff prediction, with an emphasis on simplicity and comparability to more advanced models. In particular, boosting algorithms such as Extreme Gradient Boosting (XGBoost) and Categorical Boosting (CatBoost) are examined due to their strong performance in previous hydrological studies. These models are known for handling complex, non-linear relationships and offering high accuracy with efficient computation. Their ability to manage missing and categorical data also adds to their practical advantages. The results show that XGBoost and CatBoost provide accurate and robust runoff predictions, making them promising tools for improving hydrological forecasting and supporting better water resource planning and flood risk management. Furthermore, optimization techniques, including GWO, SMA, and PSO, were incorporated to elevate forecasting accuracy. Data were carefully collected and preprocessed from reliable sources, with 80% used for training and 20% for testing. Both individual algorithms and their hybrid counterparts were evaluated, revealing XGBoost's superior performance, notably in its hybrid form with SMA, achieving an R2 value of 0.98227. The study indicates the promise of hybrid models in advancing runoff prediction, yet also emphasizes the need for further refinement in capturing peak values.
This study describes a noncontact method for measuring nanoparticle agglomeration in epoxy-based polymer nanocomposite sheets reinforced by carbon nanotubes. The method proposed in this paper utilizes active infrared thermography along with deep convolutional neural networks (DCNN). Five nanocomposite specimens were prepared with artificially generated zones of agglomeration at the centers. A thermal stimulation was provided by using a 2kW tungsten source, and infrared images at an emission frequency of 40 Hz were recorded utilizing a Fluke A253 thermal imager. The best image contrast was obtained by applying an optimal three-second thermal pulse, enabling the separation of homogeneous regions from regions of agglomerations. Agglomerations ranging between 3 mm to 8 mm in diameter were identified and classified successfully with over 92% accuracy. An eight-layer DCNN framework was implemented using three different window sizes of 41×41, 51×51, and 61×61 pixels, which was trained using 175,598 segments of images, of which 85% was made available for training, while 15% was used for evaluation. Compared to conventional thermographic inspection methods that rely on manual interpretation, the proposed approach enhances defect detection by combining active infrared imaging with deep learning, providing automated, high-resolution classification of nanoparticle agglomeration zones for more accurate and scalable evaluation of nanocomposites.
A credit approval system is a framework or process that financial institutions use to assess the creditworthiness of individuals or businesses applying for loans or credit lines. Its primary goal is to evaluate the risk associated with lending money and make informed decisions about approving or denying credit applications. Accordingly, the current paper uses a novel approach of Mouth Brooding Fish (MBF) for data classification of a credit approval system based on ensembling learning. The findings of the suggested methodology are compared with those of several methodologies, including Random Forest (RF), Gaussian Kernel (GK), Support Vector Machines (SVM), Adaptive Boosting (AdaBoost), and Multi-Layer Perceptron (MLP). Additionally, all of the input data have been standardized and mapped to 0-1 intervals. The ensemble model is built by combining the results of each machine learning (ML) model using weighted values. As a result, MBF showed noteworthy F-Score, accuracy, sensitivity, and specificity values of 91.93%, 86.95%, 97.51%, and 97.51%, respectively, in comparison to the other models that were chosen. Notably, the major innovation of this work lies in the exceptional accuracy and computational efficiency demonstrated by the proposed method, which significantly enhances the performance of data classification processes within credit approval systems by enabling faster decision-making and more reliable credit risk assessments.
This article offers an advanced argument in using a Machine Learning algorithm known as Gaussian Process Regression to predict the UCS for mixtures of soils. In particular, this study constructs special GPR prediction models dedicated to estimating UCS with a high degree of precision. The article carries out an in-depth analysis of the sensitivity of UCS to variations of influential factors. This analysis takes into account a broad dataset of different soil types from various published results of stabilization tests. Embedding two meta-heuristic algorithms, namely Adaptive Opposition Slime Mould Algorithm and Ebola Optimization Search, will further ensure accuracy from the models. These algorithms further guarantee the appropriateness of the models by considering samples of UCS for various types of soil extracted from previous test outcomes in stabilization tests. The results of this study show three models: GPEO, GPAM, and an individual GPR model. Remarkably, the GPEO model shows an R2 as high as 0.995, with a very low RMSE of 90.0. These results reflect not only the accuracy and robustness of the GPEO model but also its efficiency in soil stabilization prediction. This technique is generally very promising for UCS prediction in soil stabilization mixtures with reasonable accuracy across a wide range of engineering applications.
Credit cards, from a social standpoint, act as a catalyst for financial inclusion by providing credit to those who may not have significant resources or tangible assets. This inclusion is especially important for guaranteeing fair involvement in economic activity. Credit cards simplify online transactions, travel, and purchases. They allow for both efficiency and security in making purchases by limiting reliance on actual cash and having strict fraud prevention processes in place. Furthermore, it is important to use credit cards responsibly, as a good credit history opens doors to some of the most important financial tools: loans and mortgages. Therefore, this creates avenues for the people to have more money in their pockets and thus exploit available resources, fueling socio-economic growth and stability. The aim of this project is to predict, by the use of machine learning models, whether a certain individual will get their credit card application rejected. In this quest, the study employs random forest classification and logistic regression techniques. The investigation also involves optimization methodologies such as Pelican Optimization Algorithm (POA), Coot Optimization Algorithm (COA), and Chimp Optimization Algorithm (CHOA). The different methods tried are likely to synergize to enhance the accuracy of the prediction, which will eventually be more useful for financial purposes in forecasting credit card rejection. The results of the training phase depict that the RFPO model reached an accuracy of 0.971, proving to be a vital model in predictive skills. The RFCT model followed the GBT model closely, coming in second with an accuracy rate of 0.961. However, other models did not live up to the standard set by the two leading models. It, therefore, indicates that the best-performing models were way ahead in appropriately predicting the outcomes in relation to the investigation.
Spam emails constitute a significant percentage of email traffic and are considered a cybersecurity threat, often leading to phishing attacks, malware infections, and financial fraud. These emails, sent in bulk for commercial and malicious purposes, can bypass traditional spam filters, necessitating the development of high-accuracy models for effective detection. A major challenge in spam filtering is reducing false positives, which can lead to legitimate emails being incorrectly classified as spam, impacting users' email communication. In this study, deep learning (DL) and natural language processing (NLP) methods were employed to develop a spam detection model. Five DL-based models—Dense, CNN, LSTM, CNN-LSTM, and BERT—were evaluated. Data preprocessing included stemming, lemmatization, and text vectorization using Word2Vec to enhance feature extraction. The models were trained on a real dataset, and their accuracy was assessed using multiple evaluation indices. The findings demonstrated that, among the tested models, BERT achieved the highest accuracy (99.33%), outperforming all other approaches in spam detection. Its ability to understand contextual relationships and mitigate false positives makes it highly suitable for real-world applications. Given its computational demands, future research should focus on optimizing BERT for real-time deployment through model compression and parallel execution. Additionally, further testing on larger and more diverse datasets and implementing multilingual spam filtering capabilities will enhance its practical utility.
The Internet of Things (IoT) improves our lives by facilitating real-time communication between people and objects. Predictive analytics could turn a reactive approach into a proactive one with the rise of artificial intelligence (AI) and machine learning in the healthcare sector. As a branch of machine learning, deep learning is able to deal with large amounts of data quickly, produce valuable insights, and solves complex problems. Accurate and timely diagnosis of diseases is essential for disease prevention and early treatment. The widespread adoption of electronic medical records necessitates the development of prediction models that are more accurate to effectively harness recurrent neural network variants of deep learning. This study investigates the integration of the Internet of Things and deep learning for disease prediction and diagnosis, which highlights key advancements in data collection, preprocessing techniques, and feature extraction strategies. The potentiality of convolutional and recurrent neural networks in increasing the accuracy of diagnosis and allowing early detection of diseases is analyzed. This review shows how IoT combined with deep learning can develop an investigative disruption in the prediction and diagnosis of diseases by expanding their validity, speed, and possibility of early detection thus opening the door to new applications and effective research.
