Learner Model Research Articles

Algorithmic Bias in BERT for Response Accuracy Prediction: A Case Study for Investigating Population Validity

AbstractPretrained large language models (LLMs) have gained popularity in recent years due to their high performance in various educational tasks such as learner modeling, automated scoring, automatic item generation, and prediction. Nevertheless, LLMs are black box approaches where models are less interpretable, and they may carry human biases and prejudices because historical human data have been used for pretraining these large‐scale models. For these reasons, the prediction tasks based on LLMs require scrutiny to ensure that the prediction models are fair and unbiased. In this study, we used BERT—a pretrained encoder‐only LLM for predicting response accuracy using action sequences extracted from the 2012 PIAAC assessment. We selected three countries (i.e., Finland, Slovakia, and the United States) representing different performance levels in the overall PIAAC assessment. We found promising results for predicting response accuracy using the fine‐tuned BERT model. Additionally, we examined algorithmic bias in the prediction models trained with different countries. We found differences in model performance, suggesting that some trained models are not free from bias, and thus the models are less generalizable across countries. Our results highlighted the importance of investigating algorithmic fairness in prediction models utilizing algorithmic systems to ensure models are bias‐free.

Natural Gas Consumption Forecasting Based on Homoheterogeneous Stacking Ensemble Learning

Natural gas consumption is an important indicator of energy utilization and demand, and its scientific and high-accuracy prediction plays a key role in energy policy formulation. With the development of deep neural networks and ensemble learning, a homoheterogeneous stacking ensemble learning method is proposed for natural gas consumption forecasting. Firstly, to obtain the potential data characteristics, a nonlinear concave and convex transformation-based data dimension enhancement method is designed. Then, with the aid of a stacking ensemble learning framework, the multiscale autoregressive integrated moving average (ARIMA) and high-order fuzzy cognitive map (HFCM) methods are chosen as the base learner models, while the meta learner model is constructed via a well-designed deep neural network with long short-term memory (LSTM) cells. Finally, with the natural gas energy consumption data of national and 30 provinces (where the data of Xizang are unavailable) of China from 2000 to 2019, the numerical results show the proposed algorithm has a better forecasting performance in accuracy, robustness to noise, and sensitivity to data variations than the seven compared traditional and ensemble methods, and the corresponding model applicability rate could achieve more than 90%.

Super Learner Algorithm for Carotid Artery Disease Diagnosis: A Machine Learning Approach Leveraging Craniocervical CT Angiography.

This study introduces a machine learning (ML) approach to diagnosing carotid artery diseases, including stenosis, aneurysm, and dissection, by leveraging craniocervical computed tomography angiography (CTA) data. A meticulously curated, balanced dataset of 122 patient cases was used, ensuring reproducibility and data quality, and this is publicly accessible at (insert dataset location). The proposed method integrates a super learner model which combines adaptive boosting, gradient boosting, and random forests algorithms, achieving an accuracy of 90%. To enhance model robustness and generalization, techniques such as k-fold cross-validation, bootstrapping, data augmentation, and the synthetic minority oversampling technique (SMOTE) were applied, expanding the dataset to 1000 instances and significantly improving performance for minority classes like aneurysm and dissection. The results highlight the pivotal role of blood vessel structural analysis in diagnosing carotid artery diseases and demonstrate the superior performance of the super learner model in comparison with state-of-the-art (SOTA) methods in terms of both accuracy and robustness. This manuscript outlines the methodology, compares the results with state-of-the-art approaches, and provides insights for future research directions in applying machine learning to medical diagnostics.

Association of crop commercialization and rural households' multidimensional poverty using targeted maximum likelihood estimation

Recent evidence shows the highest prevalence of multidimensional poverty in the world. To alleviate poverty, Ethiopia is investing in stimulating agricultural commercialization for poverty reduction, driving sustainable economic growth, and creating jobs. Therefore, this study investigated the causal relationship between crop commercialization and multidimensional poverty across Ethiopian rural households using the Ethiopia Socioeconomic Survey 2018/19 data, which included 1129 crop commercialized and 1585 completely subsistence rural households. Moreover, this study also examined the causal effect of crop commercialization on the multidimensional poverty mediated by mobile phone ownership status. The study used the double robust method of the augmented inverse probability of treatment weighted (AIPTW) and targeted maximum likelihood estimation (TMLE) methods. The result showed that crop commercialization has an association with multidimensional poverty. Specifically, the TMLE with a default super learner model estimates were the most precise and suggest that the prevalence of risk of being multidimensionally poor among rural households who were crop commercialized is 12.2 % lower than for those who were completely subsistence households. Thus, it is important to have more robust policies like crop commercialization to target rural multidimensional poverty reduction. Moreover, the controlled direct effect estimation analyses showed that crop-commercialized rural households reduced the adjusted risk of multidimensional poverty by; 12 % for those who owned mobile phones and 7.8 % for those who did not. To ensure sustainable livelihoods for rural multidimensional poor households, it is necessary to have a multidisciplinary policy intervention to promote crop commercialization through the spread of mobile phones in rural areas.

Enhancing deep stochastic configuration networks: Efficient training via low-rank matrix approximation

Deep Stochastic Configuration Networks (DeepSCN) represent a class of randomized learner models distinguished by their universal approximation capabilities, rapid learning processes, and ease of implementation. These models surpass other randomized learners due to an embedded supervisory mechanism. Although the feature matrix is meticulously constructed, the predictive accuracy of DeepSCN also critically depends on the quality of the model parameters obtained from solving the corresponding system of linear equations. As the number of hidden neurons increases, the resulting expansion of the feature matrix renders matrix inversion computationally expensive, potentially unstable, and memory-intensive. This study addresses these challenges by integrating randomized algorithm-based low-rank matrix approximation to compute model parameters efficiently. A theoretical error bound for the obtained model parameters is derived, ensuring precision and accuracy. The effectiveness of the algorithm is validated on benchmark datasets for both classification and regression tasks. The results demonstrate that this integration significantly enhances the reliability and stability of DeepSCN, improving performance and scalability while greatly reducing computational demands. The method maintains stability despite substantial variations in the number of hidden neurons and mitigates overfitting caused by matrix inversion, ensuring stable and reliable training outcomes. This study advances DeepSCN training methodologies and large-scale data processing.

Ensemble learning using Gompertz function for leukemia classification

Leukemia, a life-threatening disease primarily affecting children below 15 and individuals aged 55 and above, is characterized by the overproduction of abnormal blood cells. Timely detection of leukemia plays a crucial role in significantly reducing the associated mortality rate. Given the recent developments in deep learning, the development of automated leukemia detection techniques using blood smear images has emerged as an important research area. This manuscript introduces a novel algorithm to classify microscopic blood smear images into five categories: AML, CML, ALL, CLL and normal. The study investigates the use of pre-trained convolutional neural networks powered by a fuzzy ensemble-based classifier. The fuzzy-based ensemble learning technique employs adaptable weights according to the confidence scores derived from the base learner models instead of using fixed weights for the base learner models. Gompertz function with parameters, α, β and γ are considered here. The parameters of the Gompertz function are further optimized using a grid search approach to achieve the most favorable classification outcomes. Conducting a 5-class classification, the proposed algorithm achieved an accuracy of 88.80%, surpassing the performance of traditional ensemble techniques such as average, weighted average and majority voting by a significant margin of 1.83%, 1.83% and 5.71%, respectively.

Super learner model for classifying leukemia through gene expression monitoring

Leukemia is a form of cancer that affects the bone marrow and lymphatic system, and it requires complex treatment strategies that vary with each subtype. Due to the subtle morphological differences among these types, monitoring gene expressions is crucial for accurate classification. Manual or pathological testing can be time-consuming and expensive. Therefore, data-driven methods and machine learning algorithms offer an efficient alternative for leukemia classification. This study introduced a novel super learning model that leverages heterogeneous machine learning models to analyze gene expression data and classify leukemia cells. The proposed approach incorporates an entropy-based feature importance technique to identify the gene profiles most significant to the labeling process. The strength of this super learning model lies in its final super learner, Random Forest, which effectively classifies cross-validated data from the candidate learners. Validation on a gene expression monitoring dataset demonstrates that this model outperforms other state-of-the-art models in predictive accuracy. The study contributes to the knowledge regarding the use of advanced machine learning techniques to improve the accuracy and reliability of leukemia classification using gene expression data, addressing the challenges of traditional methods that rely on clinical features and morphological examination.

Explainable Artificial Intelligence for Early Prediction of Pressure Injury Risk.

Hospital-acquired pressure injuries (HAPIs) have a major impact on patient outcomes in intensive care units (ICUs). Effective prevention relies on early and accurate risk assessment. Traditional risk-assessment tools, such as the Braden Scale, often fail to capture ICU-specific factors, limiting their predictive accuracy. Although artificial intelligence models offer improved accuracy, their "black box" nature poses a barrier to clinical adoption. To develop an artificial intelligence-based HAPI risk-assessment model enhanced with an explainable artificial intelligence dashboard to improve interpretability at both the global and individual patient levels. An explainable artificial intelligence approach was used to analyze ICU patient data from the Medical Information Mart for Intensive Care. Predictor variables were restricted to the first 48 hours after ICU admission. Various machine-learning algorithms were evaluated, culminating in an ensemble "super learner" model. The model's performance was quantified using the area under the receiver operating characteristic curve through 5-fold cross-validation. An explainer dashboard was developed (using synthetic data for patient privacy), featuring interactive visualizations for in-depth model interpretation at the global and local levels. The final sample comprised 28 395 patients with a 4.9% incidence of HAPIs. The ensemble super learner model performed well (area under curve = 0.80). The explainer dashboard provided global and patient-level interactive visualizations of model predictions, showing each variable's influence on the risk-assessment outcome. The model and its dashboard provide clinicians with a transparent, interpretable artificial intelligence-based risk-assessment system for HAPIs that may enable more effective and timely preventive interventions.

Development and Internal-external Validation of a Post-operative Mortality Risk Calculator for Pediatric Surgical Patients in Low- and Middle- Income Countries Using Machine Learning

BackgroundThe purpose of this study was to develop and validate a mortality risk algorithm for pediatric surgery patients treated at KidsOR sites in 14 low- and middle-income countries. MethodsA SuperLearner machine learning algorithm was trained to predict post-operative mortality by hospital discharge using the retrospectively and prospectively collected KidsOR database including patients treated at 20 KidsOR sites from June 2018 to June 2023. Algorithm performance was evaluated by internal-external cross-validated AUC and calibration. FindingsOf 23,905 eligible patients, 21,703 with discharge status recorded were included in the analysis, representing a post-operative mortality rate of 3.1% (671 mortality events). The candidate algorithm with the best cross-validated performance was an extreme gradient boosting model. The cross-validated AUC was 0.945 (95% CI 0.936 to 0.954) and cross-validated calibration slope and intercept were 1.01 (95% CI 0.96 to 1.06) and 0.05 (95% CI -0.10 to 0.21). For Super Learner models trained on all but one site and evaluated in the holdout site for sites with at least 25 mortality events, overall external validation AUC was 0.864 (95% CI 0.846 to 0.882) with calibration slope and intercept of 1.03 (95% CI 0.97 to 1.09) and 1.18 (95% CI 0.98 to 1.39). InterpretationThe KidsOR post-operative mortality risk algorithm had outstanding cross-validated discrimination and strong cross-validated calibration. Across all external validation sites, discrimination of Super Learner models trained on the remaining sites was excellent, though re-calibration may be necessary prior to use at new sites. This model has the potential to inform clinical practice and guide resource allocation at KidsOR sites world-wide. Type of Study and Level of EvidenceObservational Study, Level III.

Memory-Based Dynamic Bayesian Networks for Learner Modeling: Towards Early Prediction of Learners’ Performance in Computational Thinking

Artificial intelligence (AI) has demonstrated significant potential in addressing educational challenges in digital learning. Despite this potential, there are still concerns about the interpretability and trustworthiness of AI methods. Dynamic Bayesian networks (DBNs) not only provide interpretability and the ability to integrate data-driven insights with expert judgment for enhanced trustworthiness but also effectively process temporal dynamics and relationships in data, crucial for early predictive modeling tasks. This research introduces an approach for the temporal modeling of learners’ computational thinking abilities that incorporates higher-order influences of latent variables (hereafter referred to as memory of the model) and accordingly predicts learners’ performance early. Our findings on educational data from the AutoThinking game indicate that when using only first-order influences, our proposed model can predict learners’ performance early, with an 86% overall accuracy (i.e., time stamps 0, 5, and 9) and a 94% AUC (at the last time stamp) during cross-validation and 91% accuracy and 98% AUC (at the last time stamp) in a holdout test. The introduction of higher-order influences improves model accuracy in both cross-validation and holdout tests by roughly 4% and improves the AUC at timestamp 0 by roughly 2%. This suggests that integrating higher-order influences into a DBN not only potentially improves the model’s predictive accuracy during the cross-validation phase but also enhances its overall and time stamp-specific generalizability. DBNs with higher-order influences offer a trustworthy and interpretable tool for educators to foresee and support learning progression.

Design of a precise ensemble expert system for crop yield prediction using machine learning analytics

AbstractAgriculture is facing significant challenges in the development of crop yield forecasts, which are important aspects of decision‐making at the international, regional, and local levels. The area of agriculture is attracting growing attention because of increasing the demand for food supplies. To ensure future food supplies, crop yield prediction (CYP) provides the best decision‐making to assist farmers in agricultural yield forecasting efficiently. Nevertheless, CYP is a difficult endeavor because of the intricacy of the underlying mechanisms and the effect of numerous factors, including weather patterns, soil characteristics, and crop management techniques. In today's era, ensemble learning (EL) approaches have recently demonstrated significant promise for enhancing the reliability and accuracy of CYP. The success of the EL techniques depends on several facts, including how the base learner models are trained and how these are combined. This study provides important insights into the EL techniques for CYP. This paper proposes an expert system model named precise ensemble expert system for crop yield prediction (PEESCYP) to predict the best crop for agricultural land. The proposed PEESCYP model employs multiple imputation by chained equation (MICE) data imputation technique to treat the missing values of the collected dataset, the isolation forest (IF) technique for outlier detection, the ant colony optimization (ACO) technique to perform feature selection, robust scaling (RS) technique to perform data normalization, and the extra tree (ET) is used for classification to overcome the variance and overfitting problem of the single classifiers. The measurements of the proposed PEESCYP model have been collected by means of accuracy, precision, recall, and F‐1 score using a prepared dataset, which is collected from International Crops Research Institute for the Semi‐Arid Tropics (ICRISAT), and the proposed model is compared with different single‐classifier based ML models, EL models, and various existing models available in the literature. The results of this experiment underline that the proposed PEESCYP model outperforms the others.

Development of a learner model tool for predicting strength and embodied carbon for lightweight concrete production

The demand for sustainable concrete in meeting the net zero carbon target places a burden in optimizing concrete response to structural strength that satisfy acceptable embodied carbon. In most cases, a low carbon concrete is deficient in structural requirement and vice versa. This dilemma informs the need for a tool that can predict compressive strength as well as embodied carbon using the same input data. Since the use of alternative materials as cement replacement to enhance sustainability is emerging in the quest for a sustainable concrete, an optimal material that satisfy both conditions of structural integrity and sustainability is still lacking. Paucity of data in the emerging lightweight low carbon concrete using alternative materials, portends an upheave in the bias for prediction of the behaviour of lightweight low carbon concrete. This study therefore uses concrete data of lightweight low carbon concrete from laboratory experiment for the prediction of compressive strength and embodied carbon with their performance evaluated using eight machine leaning regression models. The results obtained indicates that the XG boost regression model exhibited excellent performance with a low Mean Squared Error (MSE) of 50.15, Mean absolute error(MAE) = 5.26, Mean absolute percentage error(MAPE) = 11.76 %, Explained variance score = 0.97, Root mean square error(RMSE) = 7.08 and a high R squared value of 0.96. The tool predicted compressive strength and embodied carbon for lightweight carbon concrete using multiple output regression such that the output can be limited to the yearly structural embodied carbon threshold to achieving 2050 net zero target. The developed tool when compared with concrete of similar mix ingredients performed more than 95 % in predicting concrete compressive strength and associated embodied carbon. In line with the inclusion of embodied carbon in carbon regulations of buildings in the UK as suggested by the professionals in the construction industry, the developed learner model and prediction tool has integrated concrete strength and embodied carbon to initiate a holistic approach to design and construction, balancing performance, cost, and environmental impact.

Foundation phase male teachers’ reflections in teaching of comprehensive sexuality education in the Eastern Cape province of South Africa

Children in their childhood and teenage years can acquire either healthy or unhealthy lifestyles. Without proper guidance, unhealthy lifestyles (e.g. risk-taking, substance abuse, poor family planning) may negatively affect their psychosexual development (i.e. ability to make healthier sexual choices). Male and female teachers can be role models for learners to acquire healthy sexual lifestyles. Historically, the Foundation Phase (FP) has been a highly feminine teaching space. Lately, however, the space has seen an increasing presence and participation of FP male teachers. This paper shares FP male teachers’ reflections in teaching comprehensive sexuality education (CSE) content within the Life Skills curriculum in the Eastern Cape province of South Africa. Five FP male teachers were engaged in a focus group conversation; data were thematically analysed, underpinned by a feminist post-structural lens. Although the teachers reported a restriction of FP male teachers in teaching CSE-related content, they also indicated that learners showed increased self-exposure to CSE-related content. To counter the censorship of FP male teachers in teaching CSE-related content, the study recommends explicit incorporation of feminist pedagogies to enable FP male teachers to be part of the delivery of SE content as equal stakeholders.

Forecasting Ocean Waves off the U.S. East Coast Using an Ensemble Learning Approach

Abstract This study introduces an ensemble learning model for the prediction of significant wave height and average wave period in stations along the U.S. Atlantic coast. The model utilizes the stacking method, combining three base learner models—least absolute shrinkage and selection operator (LASSO) regression, support vector machine, and multilayer perceptron—to achieve more precise and robust predictions. To train and evaluate the models, a 20-yr dataset comprising meteorological and wave data was used, enabling forecasts for significant wave height and average wave period at 1-, 3-, 6-, and 12-h intervals. The data collection involved two NOAA buoy stations situated on the U.S. Atlantic coast. The findings demonstrate that the ensemble learning model constructed through the stacking method yields significantly higher accuracy in predicting significant wave height within the specified time intervals. Moreover, the study investigates the influence of swell waves on forecasting significant wave height and average wave period. Notably, the inclusion of swell waves improves the accuracy of the 12-h forecast. Consequently, the developed ensemble model effectively estimates both significant wave height and average wave period. The ensemble model outperforms the individual models in forecasting significant wave height and average wave period. This ensemble learning model serves as a viable alternative to conventional coastal models for predicting wave parameters.

ConGCNet: Convex geometric constructive neural network for Industrial Internet of Things

The intersection of the Industrial Internet of Things (IIoT) and artificial intelligence (AI) has garnered ever-increasing attention and research interest. Nevertheless, the dilemma between the strict resource-constrained nature of IIoT devices and the extensive resource demands of AI has not yet been fully addressed with a comprehensive solution. Taking advantage of the lightweight constructive neural network (LightGCNet) in developing fast learner models for IIoT, a convex geometric constructive neural network with a low-complexity control strategy, namely, ConGCNet, is proposed in this article via convex optimization and matrix theory, which enhances the convergence rate and reduces the computational consumption in comparison with LightGCNet. Firstly, a low-complexity control strategy is proposed to reduce the computational consumption during the hidden parameters training process. Secondly, a novel output weights evaluated method based on convex optimization is proposed to guarantee the convergence rate. Finally, the universal approximation property of ConGCNet is proved by the low-complexity control strategy and convex output weights evaluated method. Simulation results, including four benchmark datasets and the real-world ore grinding process, demonstrate that ConGCNet effectively reduces computational consumption in the modelling process and improves the model’s convergence rate.

Generalisable sensor-free frustration detection in online learning environments using machine learning

Learning can generally be categorised into three domains, which include cognitive (thinking), affective (emotions or feeling) and psychomotor (physical or kinesthetic). In the learner model, acknowledging the affective aspects of learning is important for a range of learner outcomes, including motivation, persistence, and engagement. Learners’ affective states can be detected using physical (e.g. cameras) and physiological sensors (e.g., EEG) in online learning. Although these detectors demonstrate high accuracy, they raise privacy concerns for learners and present challenges in deploying them on a large scale to larger groups of students or in classroom settings. Consequently, researchers have designed an alternative method that can recognise students’ affective states at any point during online learning from their interaction with a computer-based learning platform (i.e. intelligent tutoring systems) without using any sensors. Existing sensor-free affect detectors however, are less accurate and not directly generalisable to other domains and systems. This research focuses on developing generalisable sensor-free affect detectors to identify students’ frustration during online learning using machine learning classifiers. The detectors were built by identifying minimal optimal features associated with frustration from the high-dimensional feature space through a series of experiments on a real-world students’ affective dataset, which are generalisable across various learning platforms and domains. To evaluate their accuracy and generalisability, the detectors’ performance was validated on two independent datasets collected from different educational institutions. The experimental results show that cost-sensitive Bayesian classifiers can achieve higher affect detection accuracies with a small number of generalisable features compared to other classifiers.

Large-scale evaluation of cold-start mitigation in adaptive fact learning: Knowing “what” matters more than knowing “who”

Adaptive learning systems offer a personalised digital environment that continually adjusts to the learner and the material, with the goal of maximising learning gains. Whenever such a system encounters a new learner, or when a returning learner starts studying new material, the system first has to determine the difficulty of the material for that specific learner. Failing to address this “cold-start” problem leads to suboptimal learning and potential disengagement from the system, as the system may present problems of an inappropriate difficulty or provide unhelpful feedback. In a simulation study conducted on a large educational data set from an adaptive fact learning system (about 100 million trials from almost 140 thousand learners), we predicted individual learning parameters from response data. Using these predicted parameters as starting estimates for the adaptive learning system yielded a more accurate model of learners’ memory performance than using default values. We found that predictions based on the difficulty of the fact (“what”) generally outperformed predictions based on the ability of the learner (“who”), though both contributed to better model estimates. This work extends a previous smaller-scale laboratory-based experiment in which using fact-specific predictions in a cold-start scenario improved learning outcomes. The current findings suggest that similar cold-start alleviation may be possible in real-world educational settings. The improved predictions can be harnessed to increase the efficiency of the learning system, mitigate the negative effects of a cold start, and potentially improve learning outcomes.

Rater behaviors in peer evaluation: Patterns and early detection with learner model

Peer evaluation is a common practice in team-based learning (TBL) designs, which can cover the assessment of individual or group work. However, the integrity of peer evaluation can be compromised by unserious raters—individuals who do not earnestly engage in the evaluation process. These raters may exhibit behaviors like consistently assigning the same score, rushing through evaluations, or evaluating before or long after the target presentations. This study delves into the issue of unserious peer evaluation in group presentations, with a specific focus on understanding the behavior patterns in the digital system. Utilizing evaluation behavior analysis (EBA) indicators, we identify patterns linked to unserious raters during the peer evaluation process. Meanwhile, we also connect these patterns to rating consistency and actual course performance, underscoring the significance of behavior patterns. Further, we conduct a preliminary analysis to explore the application of learner model data available before the peer evaluation starts for the early detection of unserious raters. This finding can assist teachers in providing personalized prompts and interventions before the peer evaluation stage, hence enhancing the evaluation quality through targeted interventions in a timely manner.

Factors Influencing the Satisfaction of Second Language Learners of Chinese in Online Courses.

The study aims to investigate the relationship among the key factors affecting second language learners' satisfaction with online Chinese courses and their willingness to continue utilizing them by constructing a Model of Chinese Learners' Satisfaction in Online Courses. Additionally, the influence of participants' individual differences was also explored. A total of 203 second language learners of Chinese participated in the questionnaire survey, with 5 learners further participating in interviews. Learner expectations, learner perceived quality, and learner perceived value were identified as important factors influencing learner satisfaction and willingness to continue using the online course. The results of the questionnaire survey showed that (1) learner individual differences, such as age, Chinese proficiency, weekly study duration, and offline Chinese course experience, significantly influence learner satisfaction. (2) Learner expectations have a significant positive impact on perceived quality, while perceived quality positively affects perceived value. (3) Learner satisfaction significantly influences the willingness to continue using online courses. (4) The results of the interview revealed that most learners still prefer traditional offline courses, indicating that online teaching has several shortcomings and deficiencies. Overall, this study provides some scientific and reasonable decision-making references for improving online teaching methods, aiming to enhance learner satisfaction and promote the development of online education.

Enhancing the exploitation of natural resources for green energy: An application of LSTM-based meta-model for aluminum prices forecasting

Efficient resource allocation for electric car production can be achieved by anticipating aluminum future pricing. For the electric car sector to maintain a steady supply chain, efficient production, and long-term growth, accurate forecasts are essential for policymakers to plan resource policies. The accuracy of ensemble learning models in predicting Aluminum prices was analyzed using Gene Expression programming and long short-term memory (LSTM) method. The analysis was based on monthly frequency spot settlement price data of Aluminum, copper, silver, and crude oil prices, as well as currency inflation rates of USA, China, and Peru. The study period spanned from January 1, 1994 to December 31, 2022, and also included the gross domestic product (GDP) of USA, China, and other metal minerals. Initially, we assess the level of multicollinearity between each chosen input parameter and Aluminum price volatility using Variance inflation factor-based multicollinearity. The primary innovation of this study lies in the creation of a highly precise model that utilizes the LSTM algorithm to estimate monthly Aluminum prices. This model serves as a revolutionary forecasting method. An extensive evaluation of various ensemble learning models reveals that XGBoost is the optimal model for accurately forecasting monthly Aluminum prices. The LSTM-based recurrent neural network model exhibited the lowest error, with a Mean Relative Error (MRE) ranging from 0.001 to 0.008, RMSE ranging from 54.527 to 136.044, and R2 ranging from 0.95 to 0.983. The findings demonstrated that the LSTM-based super learner models had superior performance compared to the CatBoost, LightBoost, Random Forest, XGBoost, and AdaBoost models, as evidenced by higher values of the determination coefficient (R2), value account for (VAF), and other 10 error analysis evaluators. Therefore, the utilization of an LSTM-based super learner model can serve as a dependable approach for forecasting future monthly Aluminum prices. In addition, this study introduces a logical mathematical model utilizing gene expression programming (GEP) to anticipate future Aluminum prices, which can be utilized by other researchers. Thus, we conclude that both the GEP and ensemble learning models, particularly the LSTM-based Super learner model, are appropriate for precise metal price predictions, which can benefit policymakers and assist in resource policy planning.