The integration of intelligent systems in education has become essential for addressing student attrition and enhancing academic performance. This study introduces a predictive framework for classifying students into three categories—enrolled, graduated, or dropout—using advanced machine learning techniques. The Extra Trees Classifier (ETC) serves as the baseline model, whose performance is further optimized using two recent metaheuristic algorithms: the Zebra Optimizer Algorithm (ZOA) and Golden Rush Optimization (GRO). The resulting hybrid models, designated as ETZO and ETGR, are comprehensively evaluated using multiple performance indicators, including accuracy, precision, recall, and F1-score. Experimental results reveal that the ETZO model achieves the best performance, with an overall accuracy of 92.15%, outperforming both ETGR (89.17%) and the baseline ETC model (80.4%). Moreover, the ETZO model exhibits faster convergence and lower misclassification error (7.84%), confirming its robustness and reliability in predicting diverse student outcomes. The comparative analysis highlights the significant advantage of integrating metaheuristic optimization with ensemble learning for educational data analytics. These findings demonstrate the potential of optimized machine learning systems to provide accurate predictions of student academic trajectories. By facilitating early detection of at-risk learners, the proposed framework enables educational institutions to design proactive, data-driven interventions aimed at reducing dropout rates and improving graduation success.
This research investigates the critical challenge of accurate prediction in healthcare monitoring, with a particular focus on brain stroke prediction—a task essential for timely medical intervention and improved patient outcomes. Achieving high accuracy in predictive models is crucial to reduce the risks of misdiagnosis and treatment delays, both of which can lead to severe consequences. The objective of this study is to develop a logistic regression (LR)-based model for stroke prediction, emphasizing its simplicity, interpretability, and suitability for clinical decision support. Building on a review of several machine learning (ML)-based approaches proposed for stroke prediction, this research conducts an extensive evaluation of multiple models to provide a detailed comparative analysis of their performance. Logistic regression, while computationally simple, offers advantages in transparency and reliability, making it well-suited for clinical settings where interpretability is as important as predictive performance. Through rigorous experimentation and analysis, this study identifies the strengths and limitations of various ML models and demonstrates the effectiveness of LR in balancing accuracy, interpretability, and practical application. The findings highlight that the proposed LR-based approach achieves competitive predictive performance while maintaining the clinical relevance required for healthcare adoption. By combining empirical evidence with a systematic review of existing methods, this study contributes to ongoing research in advanced stroke prediction and provides a foundation for developing robust, interpretable, and clinically applicable predictive models.
Health-wise, wild blueberries are excellent, immune-booster fruits, loaded with antioxidants and Machine Learning (ML) is one of the techniques, that can use the data for yield predictions by building a model of input variables affecting the output of a crop. Basically, ML is your all-in-one toolbox of algorithms waiting to be implemented for pattern and correlation discovery in the most influential variables, for example, bee species, weather, and yields, to name a few. In this particular case, a couple of sophisticated models namely Random Forest Regression (RFR) and Decision Tree Regression (DTR) have been brought into play in this research work to determine the best scenarios for the maximum productivity of the crop. There were three optimizers deployed to make such predictions more effective and precise: the Equilibrium Optimization Algorithm (EOA), the Seagull Optimization Algorithm (SOA), and the Coati Optimization Algorithm (COA). Consequently, the RFEO model (RFR with EOA) was at the top of the performance table with an R² of 0.989 and thus, it outperformed RFSO (RFR with SOA) (0.980) and DTEO (DTR with EOA) (0.969) by a wide margin. In turn, this indicates that EOA can be a very effective way of upgrading the prediction of Random Forest and that also brings to light RFEO as the most reliable method for forecasting wild blueberry yield. The hybrid models proposed make yield predictions more accurate for the farmers and agricultural planners that become able to efficiently allocate resources such as labor, fertilizer, and water. Besides that, accurate forecasting is a great helper of instant decision-making in terms of storage and distribution. Ultimately, these devices, through the use of which expenses can be lessened, waste can be reduced, and the conservation of wild blueberry production can be promoted, are the implements of the future.
University dropout rates, which negatively affect both the economy and the academic side of things, have been worrying educators. Those who drop out before graduation are affected both academically and economically. Their initiatives to find out which students are at risk of dropping out aim at decreasing the number of school dropouts. It is especially important given that there are very high dropout rates in online programs. Educational data mining (EDM), a promising technique, can use student data to identify potential dropouts, thus allowing teachers to be proactive and offer support. This research employs Extreme Gradient Boosting Classification (XGBC) to identify student dropout. By using the Pelican Optimization Algorithm (POA), Arithmetic Optimization Algorithm (AOA), and Sunflower Optimization Algorithm (SFOA), the study aims to enhance the performance of the XGBC model for a higher level of accuracy. The principal contribution of this paper is the implementation of explainable artificial intelligence (XAI) through SHAP-based sensitivity analysis to pinpoint the factors that influence the student's decision to drop out, thus giving a better understanding of academic, demographic, and behavioral variables. Such insights can assist higher education institutions in the effective development of intervention programs, the improvement of student retention, and the transformation of educational outcomes for the better. Through hybridization with three optimizers (tunning hyperparameters of the base models by optimization algorithms), the base model produces the following outputs: XGBC + POA (XGPO), XGBC + AOA (XGAO), and XGBC + SFOA (XGSO). With an accuracy grade of 0.977, the XGPO model performs best in the test section. With a performance of 0.955, the XGAO model ranks second. Furthermore, the XGSO model performs the least effectively, as seen by its lowest precision value of 0.890.
The study introduces a smart system for the understanding of the tourists' nonverbal behaviors through the use of machine learning and optimization techniques. Data from the Sol Cayo Guillermo Hotel were analyzed, including facial expressions, gestures, sound reactions, and spatial preferences. Data from the Sol Cayo Guillermo Hotel were analyzed, including facial expressions, gestures, sound reactions, and spatial preferences. To handle class imbalance, the Synthetic Minority Over-sampling Technique )SMOTE( method was applied, enhancing model accuracy and reliability. The dataset includes 73 original samples expanded to 300 instances using SMOTE, and consists of 23 behavioral, auditory, psychological, and demographic features. The Gaussian Mixture Model (GMM) was used for clustering tourists with similar behavioral traits, followed by classification using Random Forest (RF) and Extreme Gradient Boosting (XGBoost) algorithms and their optimized versions. Different metrics such as Accuracy, Precision, Recall, F1-score, and ROC analysis were used along with five-fold cross-validation to assess the model's performance. Results show that the optimized XGBoost model (XGPO) achieved the highest performance with an Accuracy and F1-score of 0.983, outperforming XGCO (0.967) and RFPO (0.950), while the base RF model had the lowest accuracy (0.917). These findings show the effectiveness of optimized AI models in enhancing behaviour prediction and personalized service design, which in turn can be used as a valuable source of insights to improve tourist satisfaction and management strategies in the hospitality industry.
Predicting the effect of time management on student performance using machine learning (ML) entails examining a variety of factors that influence academic achievement. The Random Forest Classification (RFC) and K Nearest Neighbor Classification (KNNC) models are used in this work, as well as 2 optimization techniques called Dandelion Optimization (DO) and Tyrannosaurus Optimization Algorithm (TOA). The 2 models and mentioned optimizers, when paired with the appropriate optimizers, develop new hybrid models that improve predicted accuracy. The findings provide insight into the link between managing time and academic success. For example, the RFTO model is in the acceptable grade, with a precision value of 0.861. This suggests that pupils who have good time management abilities are more likely to succeed academically. Furthermore, the RFDO model comes close with a precision value of 0.841, while the RFC model gets a precision value of 0.795. These data imply that enhancing time management tactics might have a considerable influence on students' academic performance. Using these models, educators and policymakers may create tailored interventions to help learners improve their time management abilities and, therefore, their academic achievement. In addition, the use of optimization techniques improves the models' predictive powers, allowing them to adapt and change in response to fresh data inputs.
Accurate and early classification of diabetes is a critical area in healthcare, as timely diagnosis plays a major role in preventing complications, supporting informed clinical decisions, and enabling more personalized treatment strategies for patients. In recent years, machine learning (ML) methods have been widely explored for medical diagnosis tasks, yet achieving consistently high and reliable accuracy for diabetes prediction remains a challenging objective due to variations in patient data, nonlinear feature interactions, and noise within clinical records. In this study, a Random Forest (RF) classifier is proposed to provide an efficient, stable, and robust solution for diabetes classification using a structured clinical dataset containing key diagnostic parameters. The performance of the model is thoroughly evaluated and benchmarked against several widely used ML algorithms, employing standard metrics such as precision, recall, accuracy, and F1-score to ensure a comprehensive assessment. Experimental results clearly demonstrate the superiority of the RF classifier, which achieved an impressive accuracy of 98% and a perfect precision score of 1.0, effectively minimizing false positives and outperforming all other baseline models included in the comparison. The strong performance highlights the RF model’s ability to capture complex feature relationships through its ensemble learning structure, leading to improved prediction reliability in clinical decision-support systems. Overall, these findings underscore the potential of the Random Forest classifier as a highly effective tool for advancing diabetes classification and improving the reliability and performance of ML-driven healthcare applications, ultimately contributing to better patient outcomes and more efficient medical practices.
Classification Information Mining (DM) strategies are highly effective tools for identifying and detecting e-banking phishing websites. This study introduces a novel approach to address the challenges and complexities associated with predicting and identifying such websites. It leverages advanced association and classification information mining techniques to develop a robust and efficient detection method. The proposed calculations identified key components and rules required for classifying phishing websites and understanding the relationships among them. A phishing case was connected to demonstrate the site's phishing preparation. The rules created from the affiliated classification show the relationship between a few critical characteristics like URL and Space Character, as well as security and encryption criteria, in the ultimate phishing location rate. The test illustrates the possibility of utilizing Acquainted Classification procedures in genuine applications and its superior execution compared to other conventional classification calculations. To predict the phishing websites, this study employed the Light Gradient Boosting Classification (LGBC) model. Additionally, three novel metaheuristic algorithms, the Prairie Dog Optimization Algorithm (PDOA), Reptile Search Algorithm (RSA), and Logic Optimization Algorithm (LOA) were utilized to enhance the performance of the LGBC model. According to the trial data, the LGLO hybrid model outperformed other algorithms. In the accuracy value at the train section, the LGLO model with the value of (0.911) has the best performance compared to the LGRS and LGPD models.
Microgrids could lower their costs in the day-ahead market through bidding in a smart distribution network. The bidding dilemma is difficult because there are many uncertainties. Despite the interdependence of gas and power prices on the balancing and day-ahead markets for a microgrid, the current study offers a two-step technique for the optimal power bids. In this instance, we looked into the interdependency of energy carriers, the impact of gas and power prices occurring simultaneously, and the impact of microgrid behavior on the power and gas bidding. The proposed network uses a two-step scenario to make bids in a microgrid. The distribution system operator receives hourly energy bids from the microgrid in the initial phase, regardless of uncertainty. The microgrid operator then assists in balancing the demand in the instantaneous market in the second phase while considering the established daily bids. By using GAMS and a scenario reduction technique, this problem is addressed as mixed-integer linear programming. The effectiveness of the suggested approach is demonstrated by numerical findings that compare scenarios for lowering the operating costs of ulti-carrier microgrids. According to the results, joint participation in the day-ahead and balancing markets reduced microgrid operation by about 2.5% compared to sole day-ahead/balancing market participation.
Text classification is a common method in textual data analysis. In this study, an attempt was made to develop a two-class classification model for gender classification using Twitter's textual data. For this purpose, 2 different approaches, namely Neural Networks (NNs) and Machine Learning (ML), were used for classification. Models based on NNs include two models, namely Convolutional Neural Network (CNN) and Convolution Long-Short Term Memory Neural Network) ConvLSTM(, which were converted from text to numerical vectors using the word2index and word embedding methods. Also, in order to increase the learning rate, the RMSprop optimizer was used in these 2 models. ML-based methods include five different models: LR, XGBoost, Support Vector Classification (SVC), RF, and AdaBoost, all of which were converted from text to numerical vectors using a one-hot encoding method. Finally, by conducting a case study based on a dataset of textual Twitter data, the accuracy of different models was investigated based on different evaluation indices. The results showed that models based on NNs, which were vectorized by word2index and word embedding methods, are more accurate in gender classification than models based on ML. Among the two investigated NN models, the XLNet-Base-Cased model is suggested to predict the gender of Twitter textual data.
The urgent need for efficient planning and strong support systems in educational settings is examined in this study, which pays particular attention to tracking students' paths and examining dropout survival rates. The project explores approaches targeted at improving dropout survival rates and utilizing predictive modeling techniques to assess critical student outcomes, including graduation and dropout. The study applies state-of-the-art machine learning techniques to establish dominant patterns and offer forecasts using a wide range of student records. Weevil Damage Optimization Algorithm, Black Widow Optimization Algorithm, and Phasor Particle Swarm Optimization form the core of 3 optimization approaches. Predictive analytics in this work is based on the Cat Boost Classifier (CATC) model, fine-tuned with the use of the presented approaches. From the careful investigation of the convergence curves obtained from the improved models, the study showed that the Cat Boost Classifier modified with the Weevil Damage Optimization Algorithm (CAWD) model was the best among all with an amazing accuracy of 0.993 after 140 iterations. In this highly critical investigation, the best option should be CAWD due to its outstanding performance. This study contributes to the discussion of evidence-based treatments and policy formulations beyond the technical contributions of involving optimization methodologies along with predictive modeling, which alludes to broadened effectiveness in strengthening student support frameworks in education. It also discusses how imperative it is to put in place proactive intervention strategies that promote students' trajectories of accomplishment and catalyze growth at the institutional level.
Readability is a critical aspect of written communication, determining how effectively readers comprehend and engage with a text. This study explores the use of machine learning techniques to predict students' reading scores, with a particular focus on Random Forest Classification (RFC) as a reliable baseline model. The research leverages a dataset of 1,000 English texts to evaluate and compare the performance of RFC, the Sooty Tern Optimization Algorithm (STOA), and the Gold Rush Optimizer (GRO) in predicting readability ratings. Key metrics, including accuracy, precision, and F1-score, are employed to assess and benchmark the models. The results demonstrate that the STOA-enhanced Random Forest model (RFST) achieves superior predictive accuracy, with an average training phase accuracy of 0.968 across different reading levels. This improvement is attributed to RFST's ability to effectively identify and utilize key textual features. In comparison, the baseline RFC and GRO-optimized Random Forest (RFGR) models achieve accuracies of 0.932 and 0.935, respectively, highlighting their limitations in capturing the intricate linguistic patterns associated with readability. This study emphasizes the importance of feature selection, particularly lexical variables, in enhancing the predictive capabilities of readability models. By integrating optimization techniques, the research offers a novel approach to improving readability prediction accuracy. The findings have significant practical implications, providing insights for developing advanced readability assessment tools to support educators, publishers, and content creators in tailoring texts to their intended audiences. This work contributes to the growing body of knowledge on readability evaluation and demonstrates the potential of machine learning and optimization methods in advancing this field.
The expansion of distributed systems has highlighted the critical importance of efficient distributed processing. Load balancing, a vital aspect of distributed processing, minimizes system response time and optimizes resource utilization, making it crucial for effective distributed resource management. Recent years have seen the development of numerous resource allocation algorithms aimed at reducing costs and energy consumption in distributed systems. However, the high time complexity of traditional resource allocation algorithms has driven the adoption of metaheuristic approaches, which offer improved efficiency and scalability. Among these, the Gray Wolf Optimization algorithm (GWO) has gained attention for its innovative approach and promising results. Comparative evaluations with algorithms like the Colonial Competitive Algorithm (CCA) provide a strong foundation for future research. Experimental results using standard datasets indicate the superior performance of GWO in resource allocation. The GWO algorithm is closer to the ideal state by more than 96% and has been able to consume the least resources. The CCA algorithm with 83% and the First-best Algorithm (FF) algorithm with 78.4% have shown a large divergence from the ideal state and have consumed more resources, which is not desirable. It achieves near-optimal allocations with fewer resources, even when dealing with large-scale datasets. GWO enhances load balancing by efficiently distributing workloads, highlighting its potential for optimizing resource allocation in distributed systems. By efficiently balancing loads and minimizing resource consumption, the GWO algorithm enhances distributed system performance, paving the way for further exploration of metaheuristic methods to ensure scalability and efficiency in complex systems.
Several stock exchanges located all over the world make up the stock market, also known as the financial market. The objective of investment in the financial market is to increase profits. Because sellers want to sell their shares at the highest price and buyers want to buy them at the lowest possible price, it is difficult to predict future market dynamics in this complex market. Predictions about the stock market have long been made using traditional methods that examine both technical and fundamental factors. Traditional prediction tools are unreliable, which has led to the rise of novel artificial intelligence-based strategies. This study aims to introduce a machine learning-based model for Shanghai Stock Exchange Index (SSE) index prediction. In this research, three optimizers—the Genetic algorithm, the Artificial Bee Colony, and the Aquila optimizer—were chosen to modify the parameters of the chosen model to assess how well Adaptive Boosting performed in stock price prediction. The study's output feature was close price forecasting of the SSE index, and the input features included open, high, low, and volume prices which were collected from January 2015 to the end of June 2023. The regression coefficient of 0.9948 indicates that the combination of this adaptive boosting and the Aquila optimizer yielded very accurate and effective results. Ultimately, the findings suggested that the hybrid model could not only significantly outperform the comparative models examined in this study, but also provide useful and practical recommendations for relevant investors, businesses, and policymakers.
This research explores the impact of yoga on Venous Clinical Severity Score (VCSS) using machine learning techniques. The study employs the Adaptive Opposition Slime Mould Algorithm (AOSM) and Mountain Gazelle Optimizer (MGO) to enhance the predictive capabilities of a Gaussian Process Classification (GPC) model. With a growing interest in non-invasive therapies, the research aims to determine how yoga influences VCSS severity over a three-month period. The study focuses on individuals diagnosed with VCSS, using machine learning to analyze complex patterns in their clinical severity ratings before and during yoga practice. The GPC model serves as a classifier, refining its predictions through the integration of AOSMA and MGO. Results indicate that in the absence of the VCSS.PRE target, the GPC model achieved a precision accuracy of 0.714, compared to 0.667 for the GPAO model and 0.796 for the GPMG model. For the VCSS.3 target, the precision accuracy improved to 0.909, 0.882, and 0.821 for the GPC, GPAO, and GPMG models, respectively. The study underscores the GPC model's adaptability in predicting VCSS changes and highlights the significance of MGO in optimizing predictive accuracy. By integrating advanced machine learning techniques, the research enhances the understanding of yoga’s potential benefits for vein health. The findings contribute to the broader application of machine learning in healthcare by demonstrating its efficacy in assessing non-invasive treatment effects.
The goal of the research is to introduce a method for differentiating children with ADHD from those without the disorder by analyzing their EEG signals while they engage in a cognitive task. This paper presents a novel technique utilizing deep learning to construct images from EEG signals. The approach involves creating two images from EEGs, utilizing both functional connectivity and spectral attributes. These constructed images are then fed into an ensemble deep framework. Three transfer learning models, namely VGGNet, Inception V3, and Inception ResNet V2, are utilized in this task, enhanced with extra layers to catch data-specific attributes. Introducing a fresh ensemble method, we aim to consolidate the outputs generated by these models through an approach centered on minimizing the error discrepancies between observed and ground-truth values. When confronted with samples yielding multiple predictions, we initiate the process by computing four distinct distance metrics - Euclidean, Hamming, Manhattan, and Cosine - for every group relative to their best possible solutions. Subsequently, the defuzzification of these distance metrics is executed through the product rules to arrive at ultimate prediction. This model classifies each sample into either the ADHD or Normal classes. Through 10-fold cross-validation method, average values of accuracy, sensitivity, specificity, F1, and false discovery rate indices achieved by suggested technique were 99.41%, 99.33%, 99.51%, 99.42%, and 0.68%, respectively, for ADHD diagnosis. These findings prove the high accuracy of proposed framework in classifying EEG signals for ADHD diagnosis compared to previous techniques.
Text classification, a significant subfield of Natural Language Processing (NLP), plays a vital role in managing and interpreting the ever-increasing volume of textual data generated by modern technology. As the integration of intelligent systems in daily life grows, the need for accurate and efficient text classification becomes more critical. This study presents a text classification approach focused on processing user commands directed at the Alexa virtual assistant. To evaluate performance, four deep learning-based models are employed: a Simple Neural Network (SNN), Bidirectional Long Short-Term Memory (Bi-LSTM), Convolutional Neural Network (CNN), and Gated Recurrent Unit (GRU). All models share a common structure, consisting of three main layers: an embedding layer with 14 dimensions as input, a unique hidden layer specific to each model, and a dense output layer with 18 units corresponding to distinct command classes. Performance metrics such as recall, precision, accuracy, and Area Under the Curve (AUC) are used to assess and compare model effectiveness. The experimental results demonstrate that the CNN model achieves the highest recall (0.86) and accuracy (0.88), making it the most effective model for classifying Alexa commands. Additionally, the SNN model records the highest precision, while CNN attains the best AUC score, highlighting its robust and consistent classification capabilities. These findings suggest that convolutional architectures are particularly well-suited for command classification tasks in virtual assistants, offering promising prospects for improving user interaction, personalization, and responsiveness in intelligent voice-controlled systems.
Reinforced concrete buildings may encounter several challenging conditions throughout their lifespan, including exposure to chloride ions. This exposure, particularly in coastal areas, may lead to a decrease in durability and degradation of the concrete structures. Utilizing experimental field findings, the application of artificial intelligence (AI) might create models to accurately predict the nonsteady state evident concrete’s chloride diffusion coefficient (Dc) over a prolonged duration. This approach can help detect the factors that have a significant impact and improve the estimation of the durability of a concrete construction. Present research showcases the use of Support Vector Regression (SVR) in forecasting the Dc of concrete under different exposure situation. The prediction models were improved by utilizing the Jellyfish search optimization (JSO), and Henry gas solubility optimization (HGSO) and were trained on a dataset including 216 data points. The findings demonstrate that the (hybrid SVR with JSO), and SHGSO (hybrid SVR with HGSO) models have significant promise in properly forecasting the Dc of concrete under different exposure situation, while maintaining suitable values of the R2. A comprehensive index includes various kinds of metrics depicts that the SJSO value of objective function (OFU) was roughly 50% smaller at 0.3769 compared to the SHGSO at 0.7284. The variance percentage between two developed models for these metrics presents that there is at least a 35% variance between the two models, where in some cases the variance gets a 95% reduction, which presents the capability and reliability of the SJSO in the prediction purposes.
Depression is a widespread mental health condition that affects millions of people globally. However, early diagnosis remains challenging due to barriers such as limited access to health care. This study examines the identification and categorization of depression utilizing tweets from the Twitter platform. Employing text analysis and classification methods, a binary classification model is developed to differentiate between depressed and non-depressed tweets. Eight diverse methods incorporating machine learning (ML) and deep learning (DL) are employed. Throughout the analysis process, the term frequency-inverse document frequency (TF-IDF), principal component analysis (PCA), and K-fold cross-validation methods are utilized. Subsequently, through a case study and various evaluation metrics, such as Accuracy, F1 Score, area under the receiver operating characteristic curve (AUC-ROC), Recall, and Precision, the accuracy of the models is assessed. The outputs reveal varying model performances across different metrics. For instance, the stochastic gradient descent (SGD) model exhibited the highest value of precision with a value of 1, while the neural network (NN) model outperformed others in terms of AUC-ROC and Recall with values of 0.9932 and 0.9718, respectively. Overall, the XGBoost (XGB) and NN models emerge as superior choices according to their performance across multiple evaluation criteria. This study highlights the potential of utilizing social media data for early depression detection, providing mental health professionals with valuable insights to identify at-risk individuals and implement timely interventions.
The stock market is a financial marketplace for buying and selling shares of companies with public listings. It is an indicator of a country's economic status, taking into consideration the activities of companies and the general state of business affairs. Stock prices are determined by supply and demand. As perilous as investment in the stock market may be, there is a possibility of its reaping huge profits in the long term. Artificial intelligence has been applied in the financial sector, the stock market included. A long-short-term memory network is one of the most common forms of artificial neural networks applied in time series analysis. Having gone through training with hundreds of input and output time steps, it provides proper forecasting of stock market values. The precision of the stock market forecasts can be improved using metaheuristic algorithms such as the Moth-flame optimizer, which will provide the best optimization of the hyperparameters for an LSTM model. The purpose of this research is to forecast stock prices by the MFO model with the LSTM network. In this research, the study has considered the Nikkei 225 dataset starting from the beginning of 2013 to the end of 2022. Conclusions: Derived conclusions based on the studied subject point towards the suggested model accuracy for stock price estimation. This proposed model gives a useful way to do analysis and project the time series value of the stock. The results of the study indicated that the suggested model is more proficient in giving high-class prediction accuracy and is suitable for the volatile stock market compared with other current techniques.
