Machine Learning Optimization Research Articles

Optimizing machine learning for space weather forecasting and event classification using modified metaheuristics

Optimizing machine learning for space weather forecasting and event classification using modified metaheuristics

Steady flow properties and spectral absorption potential of supercritical carbon dioxide nanofluids: Experimental comparison and machine learning optimization

To improve the performance of the power cycle and accelerate the realization of the carbon neutrality goal, the development and application of nanofluids are imminent. In this research, we established a supercritical nanofluid spectral absorbance prototype and prepared seven supercritical carbon dioxide nanofluids for various valuable works. The experiments indicated that titanium carbide/supercritical carbon dioxide had the most outstanding stability with the shortest stabilization time (3.2 min) and the smallest fluctuation amplitude in spectral irradiance (±6.33%). Graphene/supercritical carbon dioxide emerged as the most competitive choice in terms of deposition characteristics and spectral absorption capacity, boasting a deposition density and spectral absorbance of 8.146% and 86.39%. Titanium was the most suitable raw material for constructing the prototype. Then, a performance factor was introduced to quantify the contribution of stability and deposition characteristics, and hydroxylated multi-walled carbon nanotube/supercritical carbon dioxide was found to have the most outstanding comprehensive performance with promising applications. Finally, we combined particle swarm optimization and artificial neural networks to develop an improved machine learning optimization model. Compared to the original prediction model, the root mean square error and mean absolute error of this model were reduced by 64.02% and 66.8%. Significantly, the spectral irradiance through the supercritical carbon dioxide nanofluid was reduced by 68.21% compared to the base working fluid, providing guidelines for improving the performance of the power cycle.

Accelerating Battery Electrolyte Design Process Via Closed-Loop Optimization of Simulations and Experiments

Designing and discovering new material combinations for battery applications has been and remains to be a slow and expensive process. It is a challenging task to develop battery electrolytes that are safe, high-performant, and cost-effective, as there are many design parameters to consider. There are many recent efforts to expedite such a process, ranging from the use of machine learning optimization techniques and lab automation to utilizing computer simulations.We propose an automated, closed-loop optimization strategy to design electrolytes by integrating machine learning-based optimization algorithms, autonomous robotics, and molecular simulations. In particular, we demonstrate the application of our approach for designing room-temperature ionic liquids (RTILs) for Li-ion batteries, which consist of multiple salts, cations, and anions. The composition space is too vast to explore via a conventional approach. Thus, we use intelligent decision-making algorithms for selecting which composition to test next, which is carried out by the in-house robotic platform. The molecular simulations are also used to provide an in-depth understanding of the experimentally observed behavior, and the simulation results are also fed into the optimization algorithm. The presented approach is designed to be general, so it can be applied to other electrochemical applications with minimal modifications to the algorithms or the experimental setup.

Synergistic D-Amino Acids Based Antimicrobial Cocktails Formulated via High-Throughput Screening and Machine Learning.

Antimicrobial resistance (AMR) from pathogenic bacterial biofilms has become a global health issue while developing novel antimicrobials is inefficient and costly. Combining existing multiple drugs with enhanced efficacy and/or reduced toxicity may be a promising approach to treat AMR. D-amino acids mixtures coupled with antibiotics can provide new therapies for drug-resistance infection with reduced toxicity by lower drug dosage requirements. However, iterative trial-and-error experiments are not tenable to prioritize credible drug formulations, owing to the extremely large number of possible combinations. Herein, a new avenue is provide to accelerate the exploration of desirable antimicrobial formulations via high-throughput screening and machine learning optimization. Such an intelligent method can navigate the large search space and rapidly identify the D-amino acid mixtures with the highest anti-biofilm efficiency and also the synergisms between D-amino acid mixtures and antibiotics. The optimized drug cocktails exhibit high antimicrobial efficacy while remaining non-toxic, which is demonstrated not only from in vitro assessments but also the first in vivo study using a lung infection mouse model.

Applying Learning and Self-Adaptation to Dynamic Scheduling

Real-world production scheduling scenarios are often not discrete, separable, iterative tasks but rather dynamic processes where both external (e.g., new orders, delivery shortages) and internal (e.g., machine breakdown, timing uncertainties, human interaction) influencing factors gradually or abruptly impact the production system. Solutions to these problems are often very specific to the application case or rely on simple problem formulations with known and stable parameters. This work presents a dynamic scheduling scenario for a production setup where little information about the system is known a priori. Instead of fully specifying all relevant problem data, the timing and batching behavior of machines are learned by a machine learning ensemble during operation. We demonstrate how a meta-heuristic optimization algorithm can utilize these models to tackle this dynamic optimization problem, compare the dynamic performance of a set of established construction heuristics and meta-heuristics and showcase how models and optimizers interact. The results obtained through an empirical study indicate that the interaction between optimization algorithm and machine learning models, as well as the real-time performance of the overall optimization system, can impact the performance of the production system. Especially in high-load situations, the dynamic algorithms that utilize solutions from previous problem epochs outperform the restarting construction heuristics by up to ~24%.

Analyzing the impact of feature selection methods on machine learning algorithms for heart disease prediction

The present study examines the role of feature selection methods in optimizing machine learning algorithms for predicting heart disease. The Cleveland Heart disease dataset with sixteen feature selection techniques in three categories of filter, wrapper, and evolutionary were used. Then seven algorithms Bayes net, Naïve Bayes (BN), multivariate linear model (MLM), Support Vector Machine (SVM), logit boost, j48, and Random Forest were applied to identify the best models for heart disease prediction. Precision, F-measure, Specificity, Accuracy, Sensitivity, ROC area, and PRC were measured to compare feature selection methods' effect on prediction algorithms. The results demonstrate that feature selection resulted in significant improvements in model performance in some methods (e.g., j48), whereas it led to a decrease in model performance in other models (e.g. MLP, RF). SVM-based filtering methods have a best-fit accuracy of 85.5. In fact, in a best-case scenario, filtering methods result in + 2.3 model accuracy. SVM-CFS/information gain/Symmetrical uncertainty methods have the highest improvement in this index. The filter feature selection methods with the highest number of features selected outperformed other methods in terms of models' ACC, Precision, and F-measures. However, wrapper-based and evolutionary algorithms improved models' performance from sensitivity and specificity points of view.

Multilocus Distance-Regulated Sensor Array for Recognition of Polyphenols via Machine Learning and Indicator Displacement Assay.

Developing new strategies to construct sensor arrays that can effectively distinguish multiple natural components with similar structures in mixtures is an exceptionally challenging task. Here, we propose a new multilocus distance-modulated indicator displacement assay (IDA) strategy for constructing a sensor array, incorporating machine learning optimization to identify polyphenols. An 8-element array, comprising two fluorophores and their six dynamic covalent complexes (C1-C6) formed by pairing two fluorophores with three distinct distance-regulated quenchers, has been constructed. Polyphenols with diverse spatial arrangements and combinatorial forms compete with the fluorophores by forming pseudocycles with quenchers within the complexes, leading to varying degrees of fluorescence recovery. The array accurately and effectively distinguished four tea polyphenols and 16 tea varieties, thereby demonstrating the broad applicability of the multilocus distance-modulated IDA array in detecting polyhydroxy foods and natural medicines.

Optimized Machine learning techniques Enable prediction of organic dyes photophysical Properties: Absorption Wavelengths, emission Wavelengths, and quantum yields

Applications of organic dyes, ranging from basic research to industry, are functions of their photophysical properties. Two important aspects— (1) knowledge of the photophysical properties of existing dyes long before real applications and (2) discovery of new organic dyes with desired photophysical properties for either upgradation of existing or development of new applications—are needed to be addressed. These two cases are coupled together with the common goal of estimating photophysical properties with high accuracy at the minimum cost of time and money long before the hard-core laboratory experiment. For this purpose, machine learning-based techniques are the most suitable approach. In this study, we used optimized machine-learning techniques to assess a dataset of 3066 organic dyes, which were evaluated using three evaluation parameters: Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and the coefficient of determination (R2). The Quadratic Support Vector Machine (QSVM) was the best predictive model for RMSE-16.614, MAE-10.837, and R2-0.961 for absorption wavelengths and RMSE-23.636, MAE-16.278, and R2-0.929 for emission wavelengths. These R2 values are 0.7% and 0.4% greater than the Gradient Boost Regression Tree (GBRT) model's recently reported values of 0.954 and 0.925 for absorption and emission wavelengths, respectively. Furthermore, we estimated the quantum yield and found that the Coarse Gaussian Support Vector Machine (CGSVM) outperformed all examined models. For more validation of these models, we compared the predicted results with the experimental results of selective dyes. The proposed automated approach can be used for predicting photophysical properties without much computer programming knowledge.

Exploring the association between two groups of metals with potentially opposing renal effects and renal function in middle-aged and older adults: Evidence from an explainable machine learning method

BackgroundMachine learning models have promising applications in capturing the complex relationship between mixtures of exposures and outcomes. ObjectiveOur study aimed at introducing an explainable machine learning (EML) model to assess the association between metal mixtures with potentially opposing renal effects and renal function in middle-aged and older adults. MethodsThis study extracted data from two cycle years of the National Health and Nutrition Examination Survey (NHANES). Participants aged 45 years or older with complete data on six metals (lead, cadmium, manganese, mercury, and selenium) and related covariates were enrolled. The EML model was developed by the optimized machine learning model together with Shapley Additive exPlanations (SHAP) to assess the chronic kidney disease (CKD) risk with metal mixtures. The results from EML were further compared in detail with multiple logistic regression (MLR) and Bayesian kernel machine regression (BKMR). ResultsAfter adjusting for included covariates, MLR pointed out the lead and arsenic were generally positively associated with CKD, but manganese had a negative association. In the BKMR analysis, each metal was found to have a non-linear association with the risk of CKD, and interactions can exist between metals, especially for arsenic and lead. The EML ranked the feature importance: lead, manganese, arsenic and selenium were close behind in importance after gender, age or BMI for participants with CKD. Strong interactions between mercury and lead, manganese and cadmium and arsenic and manganese were identified by partial dependence plot (PDP) of SHAP and bivariate exposure-response effect plots of BKMR. The EML model determined the “trigger point” at which the risk of CKD abruptly changed. ConclusionCo-exposure to metals with different nephrotoxicity could have different joint association with renal function, and EML can be a powerful method for studying complex exposure mixtures.

Machine Learning Algorithms for Slow Fashion Consumer Prediction: Theoretical and Managerial Implications

Purpose: To compare, propose, and discuss the implications of five machine learning algorithms for predicting Slow fashion consumer profiles. Methodology/approach: We use the Python programming language to build the models with scikit-learn libraries. We tested the potential of five algorithms to correct classifier Slow fashion consumers: I) extremely randomized trees, II) random forest, III) support vector machine, IV) gradient boosting Tree, and V) naïve bayes. Originality/Relevance: This paper's originality lies in its combination of sustainability concerns, consumer behavior analysis, and machine learning techniques. It addresses a critical issue in the fashion industry and offers practical implications that can be beneficial for companies seeking to align their practices with Slow fashion principles. This interdisciplinary approach makes it a relevant contribution to both academia and industry. Key findings: The performance metrics revealed satisfactory values for all algorithms. Nevertheless, the support vector machine presents a better precision (96%) on the dataset for Slow fashion consumer profiling, while random forest performs the worst (87%). Theoretical/methodological contributions: We understood that the model can be helpful for companies that wish to adopt more targeted and practical approaches in the context of Slow fashion, allowing them to make more informed and strategic decisions. Therefore, these insights can guide future research in optimizing machine learning applications for consumer behavior analysis and provide valuable guidance for fashion marketers seeking to enhance their targeting and engagement strategies.

Enhancing Neural Network with Particle Swarm: A House Price Prediction Case

This study delves into the utilization of the Particle Swarm Optimization (PSO) algorithm to discover the best parameters of Long Short-Term Memory (LSTM) model with the aim of housing price prediction. Employing the Ames Housing dataset, our investigation underscores the remarkable efficacy of PSO in mitigating prediction errors within the LSTM framework, thereby enhancing overall predictive performance. The symbiotic integration of PSO with LSTM facilitates the discernment of intricate data patterns, particularly in scenarios involving high-dimensional and nonlinear optimization conundrums. The findings illuminate PSO's ability to expediently traverse and identify global optima. The outcomes unequivocally underscore the PSO-LSTM model's superior attributes, characterized by accelerated convergence, diminished errors, and exceptional proficiency. In summation, this investigation underscores PSO's promise as a potent technique for optimizing machine learning models, notably exemplified by its integration with the LSTM framework.

Machine Learning Algorithms for Slow Fashion Consumer Prediction: Theoretical and Managerial Implications

Purpose: To compare, propose, and discuss the implications of five machine learning algorithms for predicting Slow fashion consumer profiles. Methodology/approach: We use the Python programming language to build the models with scikit-learn libraries. We tested the potential of five algorithms to correct classifier Slow fashion consumers: I) extremely randomized trees, II) random forest, III) support vector machine, IV) gradient boosting Tree, and V) naïve bayes. Originality/Relevance: This paper's originality lies in its combination of sustainability concerns, consumer behavior analysis, and machine learning techniques. It addresses a critical issue in the fashion industry and offers practical implications that can be beneficial for companies seeking to align their practices with Slow fashion principles. This interdisciplinary approach makes it a relevant contribution to both academia and industry. Key findings: The performance metrics revealed satisfactory values for all algorithms. Nevertheless, the support vector machine presents a better precision (96%) on the dataset for Slow fashion consumer profiling, while random forest performs the worst (87%). Theoretical/methodological contributions: We understood that the model can be helpful for companies that wish to adopt more targeted and practical approaches in the context of Slow fashion, allowing them to make more informed and strategic decisions. Therefore, these insights can guide future research in optimizing machine learning applications for consumer behavior analysis and provide valuable guidance for fashion marketers seeking to enhance their targeting and engagement strategies.

BRDF modeling and optimization of a target surface based on the gradient descent algorithm

Addressing the current challenges in modeling and optimizing the bidirectional reflectance distribution function (BRDF) for the target surface, an improved six-parameter semi-empirical model is proposed based on an existing five-parameter semi-empirical model. In comparison with the original five-parameter model, the new, to the best of our knowledge, model considers reciprocity, and the results demonstrate that as the incident angle increases, the fitting accuracy of the six parameters gradually surpasses that of the five parameters. Additionally, this paper employs a machine learning optimization algorithm, namely, the gradient descent method, for optimizing the BRDF. The algorithm was comprehensively compared with other optimization methods, revealing that for the same dataset, the gradient descent method exhibited the smallest fitting errors. Subsequently, utilizing this algorithm for fitting experimental data resulted in errors consistently within 3%, confirming the reliability and accuracy of this optimization algorithm.

Progress in research of models for predicting the risk of readmission and mortality among patients with acute heart failure

Heart failure is a serious and end-stage status of various heart diseases, characterized by comparatively high rate of readmission and mortality, and has become an important public health issue. The risk of readmission and mortality following discharge of an index hospitalization are key indicators to evaluate the quality of medical care among patients with acute heart failure. Therefore, it is important to carry out risk prediction research for patients with acute heart failure, quantify the disease risk, perform risk stratification, optimize clinical decision-making, elevate patients' quality of life and prognosis, and comprehensively improve the medical quality of acute heart failure. During the past 20 years, foreign researchers have developed dozens of models to predict the risk of acute heart failure readmission and mortality, and Chinese researchers have also developed up to 10 models applicable to the Chinese population. However, there is no recommended risk prediction model for acute heart failure in current clinical guidelines across China. In this report, we aim to introduce the major models for predicting the risk of acute heart failure readmission and mortality from home and abroad, focus on putting forward limitations of established models, and initiating potential directions for future studies from the following aspects: integrate multi-source data, mine emerging biomarkers, establish polygenic risk scores, optimize machine learning methods, promote flexible adjustment, and broaden approaches that applicable for various scenarios. Accordingly, this study will help facilitate domestic research in predicting the risk of readmission and mortality among patients hospitalized for acute heart failure.

RETRACTED ARTICLE: Optical bio sensor based cancer cell detection using optimized machine learning model with quantum computing

RETRACTED ARTICLE: Optical bio sensor based cancer cell detection using optimized machine learning model with quantum computing

InfrastructureModels: Composable Multi-infrastructure Optimization in Julia

In recent years, there has been an increasing need to understand the complex interdependencies between critical infrastructure systems, for example, electric power, natural gas, and potable water. Whereas open-source and commercial tools for the independent simulation of these systems are well established, frameworks for cosimulation with other systems are nascent and tools for co-optimization are scarce—the major challenge being the hidden combinatorics that arise when connecting multiple-infrastructure system models. Building toward a comprehensive solution for modeling interdependent infrastructure systems, this work presents InfrastructureModels, an extensible, open-source mathematical programming framework for co-optimizing multiple interdependent infrastructures. This work provides new insights into methods and programming abstractions that make state-of-the-art independent infrastructure models composable with minimal additional effort. To that end, this paper presents the design of the InfrastructureModels framework, documents key components of the software’s implementation, and demonstrates its effectiveness with three case studies on canonical co-optimization tasks arising in interdependent infrastructure systems. History: Accepted by Ted Ralphs, Area Editor for Software Tools. Funding: The work was funded by Los Alamos National Laboratory’s Directed Research and Development project “The Optimization of Machine Learning: Imposing Requirements on Artificial Intelligence” and the U.S. Department of Energy’s Office of Electricity Advanced Grid Modeling projects “Joint Power System and Natural Gas Pipeline Optimal Expansion Planning” and “Coordinated Planning and Operation of Water and Power Infrastructures for Increased Resilience and Reliability.” This work was carried out under the U.S. DOE contract no. [DE-AC52-06NA25396]. Supplemental Material: The software that supports the findings of this study is available within the paper and its Supplemental Information ( https://pubsonline.informs.org/doi/suppl/10.1287/ijoc.2022.0118 ) as well as from the IJOC GitHub software repository ( https://github.com/INFORMSJoC/2022.0118 ). The complete IJOC Software and Data Repository is available at https://informsjoc.github.io/ .

Prediction of the Productivity Ratio of Perforated Wells Using Least Squares Support Vector Machine with Particle Swarm Optimization

The productivity ratio is a vital metric for assessing the efficiency of perforated completions. Accurate and rapid prediction of this ratio is essential for optimizing the perforation design. In this study, we propose a novel approach that combines three-dimensional finite element numerical simulation and machine learning techniques to predict the productivity ratio of perforated wells. Initially, we obtain the productivity ratio of perforated wells under various perforation parameters using three-dimensional finite element numerical simulation. This generates a sample set for machine learning. Subsequently, we employ the least squares support vector machine (LSSVM) algorithm to establish a prediction model for the productivity ratio of perforated wells. To optimize the parameters of the LSSVM algorithm, we utilize the particle swarm optimization (PSO) algorithm. We compare our proposed PSO-LSSVM model with that established based on other parameter optimization methods and machine learning algorithms, such as Grid search-LSSVM, PSO-ANN, and PSO-RF. Our results demonstrate that the PSO-LSSVM model exhibits rapid convergence, high prediction accuracy, and strong generalization ability in predicting the productivity ratio of perforated wells. This research provides a valuable reference and guidance for optimizing perforation design. Additionally, it offers new insights into predicting the productivity of complex completions.

Comparative analysis of machine learning algorithms on myoelectric signal from intact and transradial amputated limbs

<p>Control strategies of smart hand prosthesis-based myoelectric signals in<br />recent years don't provide the patients with the sensation of biological<br />control of prostheses hand fingers. Therefore, in current work<br />hyperparameters optimization in machine learning algorithm and hand<br />gesture recognition techniques were applied to the myoelectric signal-based<br />on residual muscles contraction of the amputees corresponding to intact<br />forearm limb movement to improve their biological control. In this paper,<br />myoelectric signals are extracted using the MYO armband to recognize ten<br />gestures from ten volunteers (healthy and transradial amputation) on the<br />forearm, thereafter the noise of myoelectric signals using a notch filter (NF)<br />is removed. The proposed classification system involved two machine<br />learning algorithms: (1) the decision tree (DT), tri-layered neural network<br />(TLNN), k-nearest-neighbor (KNN), support vector machine (SVM) and<br />ensemble boosted tree (EBT) classifiers. (2) the optimized machine learning<br />classifiers, i.e., OKNN, OSVM, OEBT with optical diffraction tomography<br />(ODT) and ommatidia detecting algorithm (ODA). The experimental results<br />of classifiers comparison pointed out an algorithm that outperformed with<br />high accuracy is OEBT closely followed by OKNN achieves an accuracy of<br />97.8% and 97.1% for intact forearm limb, while for transradial amputation<br />with an accuracy of 91.9% and 91.4%, respectively.</p>

Optimal Data Splitting in Distributed Optimization for Machine Learning

Optimal Data Splitting in Distributed Optimization for Machine Learning

Predicting slope safety using an optimized machine learning model

The hazards and consequences of slope collapse can be reduced by obtaining a reliable and accurate prediction of slope safety, hence, developing effective tools for foreseeing their occurrence is crucial. This research aims to develop a state-of-the-art hybrid machine learning approach to estimate the factor of safety (FOS) of earth slopes as precisely as possible. The current research’s contribution to the body of knowledge is multifold. In the first step, a powerful optimization approach based on the artificial electric field algorithm (AEFA), namely the global-best artificial electric field algorithm (GBAEF), is developed and verified using a number of benchmark functions. The aim of the following step is to utilize the machine learning technique of support vector regression (SVR) to develop a predictive model to estimate the slope’s safety factor (FOS). Finally, the proposed GBAEF is employed to enhance the performance of the SVR model by appropriately adjusting the hyper-parameters of the SVR model. The model implements 153 data sets, including six input parameters and one output parameter (FOS) collected from the literature. The outcomes show that implementing efficient optimization algorithms to adjust the hyper-parameters of the SVR model can greatly enhance prediction accuracy. A case study of earth slope from Chamoli District, Uttarakhand is used to compare the proposed hybrid model to traditional slope stability techniques. According to experimental findings, the new hybrid AI model has improved FOS prediction accuracy by about 7% when compared to other forecasting models. The outcomes also show that the SVR optimized with GBAEF performs wonderfully in the disciplines of training and testing, with a maximum R2 of 0.9633 and 0.9242, respectively, which depicts the significant connection between observed and anticipated FOS.