Hyperparameter Combinations Research Articles

Pseudo-Siamese network combined with label-free Raman spectroscopy for the quantification of mixed trace amounts of antibiotics in human milk: A feasibility study

The utilization of antibiotics is prevalent among lactating mothers. Hence, the rapid determination of trace amounts of antibiotics in human milk is crucial for ensuring the healthy development of infants. In this study, we constructed a human milk system containing residual doxycycline (DXC) and/or tetracycline (TC). Machine learning models and clustering algorithms were applied to classify and predict deficient concentrations of single and mixed antibiotics via label-free SERS spectra. The experimental results demonstrate that the CNN model has a recognition accuracy of 98.85% across optimal hyperparameter combinations. Furthermore, we employed Independent Component Analysis (ICA) and the pseudo-Siamese Convolutional Neural Network (pSCNN) to quantify the ratios of individual antibiotics in mixed human milk samples. Integrating the SERS technique with machine learning algorithms shows significant potential for rapid discrimination and precise quantification of single and mixed antibiotics at deficient concentrations in human milk.

Machine learning methods for financial forecasting and trading profitability: Evidence during the Russia–Ukraine war

PurposeThis study aims to evaluate the effectiveness of machine learning models to yield profitability over the market benchmark, notably in periods of systemic instability, such as the ongoing war between Russia and Ukraine.Design/methodology/approachThis study made computational experiments using support vector machine (SVM) classifiers to predict stock price movements for three financial markets and construct profitable trading strategies to subsidize investors’ decision-making.FindingsOn average, machine learning models outperformed the market benchmarks during the more volatile period of the Russia–Ukraine war, but not during the period before the conflict. Moreover, the hyperparameter combinations for which the profitability is superior were found to be highly sensitive to small variations during the model training process.Practical implicationsInvestors should proceed with caution when applying machine learning models for stock price forecasting and trading recommendations, as their superior performance for volatile periods – in terms of generating abnormal gains over the market – was not observed for a period of relative stability in the economy.Originality/valueThis paper’s approach to search for financial strategies that succeed in outperforming the market provides empirical evidence about the effectiveness of state-of-the-art machine learning techniques before and after the conflict deflagration, which is of potential value for researchers in quantitative finance and market professionals who operate in the financial segment.

Optimal multivariate mixture: a genetic algorithm approach

AbstractThe Optimal Multivariate Mixture Problem (OMMP) consists of finding an optimal mixture which, starting from a set of elements (items) described by a set of variables (features), is as close as possible to an ideal solution. This problem has numerous applications spanning various fields, including food science, agriculture, chemistry, materials science, medicine, and pharmaceuticals. The OMMP is a class of optimization problems that can be addressed using traditional Operations Research (OR) approaches. However, it can also be effectively tackled using meta-heuristic techniques within Artificial Intelligence (AI). This paper aims to present an Artificial Intelligence perspective. It proposes a Genetic Algorithm (GA) for Optimal Multivariate Mixture (GA-OMM), a novel improved version of a GA whose modified genetic operators prove to improve the exploration efficiency. Here, the algorithm is described in its general framework, and a test case 8-items 5-features is conducted to evaluate efficiency by exploring various combinations of hyperparameters. Test cases are also set up for the previous version, as well as a linear programming (LP) approach. The data experiments indicate that the proposed GA is efficient, converges towards the global optimum, consistently outperforms its predecessor, and delivers highly competitive results. In particular, GA-OMM shows an average fitness of GA-OMMP/LP and standard deviation with an order of magnitude ranging between $$10^{-8}$$ 10 - 8 to $$10^{-4}$$ 10 - 4 . Moreover, it consistently outperforms its predecessor, which exhibits similar values around $$10^{-3}$$ 10 - 3

In-Depth Exploration and Comparison of Machine Learning Performances for Early-Stage Diabetes Risk Prediction

Abstract — Diabetes mellitus is distinguished by an inability of the human system to produce insulin on an ongoing basis, as well as by the inefficient utilization of the insulin hormone, resulting in an elevated level of blood glucose. Global diabetes rates have nearly doubled since 1980, reaching 9.3% among adults. Alarmingly, of the 463 million individuals with diabetes, 50.1% are unaware of their condition. Indonesia ranks seventh globally with 10.7 million diabetes cases. In 2019, it was fifth globally for adults (20–79 years) with undiagnosed diabetes. This silent epidemic demands urgent attention and comprehensive strategies for early detection and management. In recent years, researchers have increasingly studied machine learning for early diabetes recognition. In this study, we aim to predict early-stage diabetes risk by utilizing 16 health condition features. We explore 12 distinct machine learning algorithms, applying a hyperparameter grid to tune each algorithm. This involves systematically testing combinations of hyperparameters to identify the optimal settings for achieving the most accurate and reliable predictive models. The results indicate that the Light GBM algorithm achieved the highest accuracy of 0.9692. By contrast, the logistic regression and Naive Bayes algorithms demonstrated the lowest performance, each with an accuracy of 0.8923. The implications of these results underline the capability of employing machine learning algorithms to precisely and effectively detect individuals susceptible to diabetes, enabling the implementation of individualized healthcare approaches.

다중 센서 데이터를 활용한 오토인코더 기반 화재감지 모델

Large-scale fires and their consequential damages are becoming increasingly common, but confidence in fire detection systems is waning. Recently, widely-used chemical fire detectors frequently generate lots of false alarms, while video-based deep learning fire detection is hampered by its time-consuming and expensive nature. To tackle these issues, this study proposes a fire detection model utilizing an autoencoder approach. The objective is to minimize false alarms while achieving swift and precise fire detection. The proposed model, employing an autoencoder methodology, can exclusively learn from normal data without the need for fire-related data, thus enhancing its adaptability to diverse environments. By amalgamating data from five distinct sensors, it facilitates rapid and accurate fire detection. Through experiments with various hyperparameter combinations, the proposed model demonstrated that out of 14 scenarios, only one encountered false alarm issues. Experimental results underscore its potential to curtail fire-related losses and bolster the reliability of fire detection systems.

Prediction of hydrogen leakage location and intensity in hydrogen refueling stations based on deep learning

The common risk of high-pressure hydrogen leakage seriously affects the safety of hydrogen refueling stations (HRS). In this study, we have developed a hybrid model (WD-KNN-CNN) based on wavelet denoising (WD) and deep learning to predict the hydrogen leakage location and intensity in HRS. The raw data were calculated using validated Computational Fluid Dynamics (CFD), and the WD-processed raw data were utilized as input data for the hybrid model. The K-nearest Neighbors (KNN) is employed to identify the leak location, while a two-dimensional Convolutional Neural Network (CNN) is utilized to extract temporal and spatial distribution features from the concentration sequence for intensity prediction. Bayesian optimization is applied to determine the optimal hyperparameter combination for the model. The results demonstrate that the model achieves a prediction accuracy of 99.14% for leak location and 97.42% for intensity level. Furthermore, through the reduction of sensor quantity, we tested the model's performance and found that utilizing 14 sensors as input resulted in 91.3% accuracy for leak location and 90.8% for intensity level prediction. This hybrid model provides a method for safety monitoring in HRS and offers guidance for personnel response during leak incidents.

Long-term monitoring chlorophyll-a concentration using HJ-1 A/B imagery and machine learning algorithms in typical lakes, a cold semi-arid region.

Chlorophyll a (Chl-a) in lakes serves as an effective marker for assessing algal biomass and the nutritional level of lakes, and its observation is feasible through remote sensing methods. HJ-1 (Huanjing-1) satellite, deployed in 2008, incorporates a CCD capable of a 30 m resolution and has a revisit interval of 2 days, rendering it a superb choice or supplemental sensor for monitoring trophic state of lakes. For effective long-term and regional-scale mapping, both the imagery and the evaluation of machine learning algorithms are essential. The several typical machine learning algorithms, i.e., Support Vector Regression (SVR), Gradient Boosting Decision Trees (GBDT), XGBoost (XGB), Random Forest (RF), K-Nearest Neighbor (KNN), Kernel Ridge Regression (KRR), and Multi-Layer Perception Network (MLP), were developed using our in-situ measured Chl-a. A cross-validation grid to identify the most effective hyperparameter combinations for each algorithm was used, as well as the selected optimal superparameter combinations. In Chl-a mapping of three typical lakes, the R2 of GBDT, XGB, RF, and KRR all reached 0.90, while XGB algorithm also exhibited stable performance with the smallest error (RMSE = 3.11 μg/L). Adjustments were made to align the Chl-a spatial-temporal patterns with past data, utilizing HJ1-A/B CCD images mapping through XGB algorithm, which demonstrates its stability. Our results highlight the considerable effectiveness and utility of HJ-1 A/B CCD imagery for evaluation and monitoring trophic state of lakes in a cold arid region, providing the application cases contribute to the ongoing efforts to monitor water qualities.

Tugboat Scheduling Method Based on the NRPER-DDPG Algorithm: An Integrated DDPG Algorithm with Prioritized Experience Replay and Noise Reduction

The scheduling of harbor tugboats is a crucial task in port operations, aiming to optimize resource allocation and reduce operational costs, including fuel consumption of tugboats and the time cost of vessels waiting for operations. Due to the complexity of the port environment, traditional scheduling methods, often based on experience and practice, lack scientific and systematic decision support, making it difficult to cope with real-time changes in vessel dynamics and environmental factors. This often leads to scheduling delays and resource waste. To address this issue, this study proposes a mathematical model based on fuzzy programming, accounting for the uncertainty of the arrival time of target vessels. Additionally, we introduce the NRPER-DDPG algorithm (DDPG Algorithm with Prioritized Experience Replay and Noise Reduction), which combines a prioritized replay mechanism with a decaying noise strategy based on the DDPG algorithm. This approach optimizes the time for tugboats to reach the task location as a continuous action space, aiming to minimize the total system cost and improve scheduling efficiency. To verify the effectiveness of the mathematical model and algorithm, this study conducted experimental validation. Firstly, the optimal algorithm hyperparameter combinations were adjusted through random examples to ensure the stability and reliability of the algorithm. Subsequently, large-scale examples and actual port cases were used to further verify the performance advantages of the algorithm in practical applications. Experimental results demonstrate that the proposed mathematical model and algorithm significantly reduce system costs and improve scheduling efficiency, providing new insights and methods for the sustainable development of port operations.

Ensemble Residual Network with Iterative Randomized Hyperparameter Optimization for Colorectal Cancer Classification

The analysis of WSI images is widely acknowledged as a method, for identifying stages of cancer and evaluating the spread of cancer cells in tissues. In histopathology image analysis deep learning models are gaining increasing importance. To enhance the effectiveness of these models it is crucial to train and fine-tune Convolutional Neural Network algorithms by adjusting hyperparameters like batch size, convolution depth, and learning rate (LR). However, determining the hyperparameters can be challenging as they significantly impact model performance. This study examines how hyperparameters influence cancer classification, in histopathology images using the CNN architecture. A method called iterative randomized hyperparameter optimization is proposed, which gradually reduces variations over time by adjusting parameters after each network layer. The combination of hyperparameters is applied to version 1 of ResNet18, ResNet50, and ResNet101 models and version 2 of ResNet50, ResNet101, and ResNet151. The results are then combined using the Adaptive Boosting algorithm. The results are quite promising on ensemble version 1 residual networks, achieving an accuracy of 98.7% when tested on nine tissue classes.

Hyperparameter optimization approaches to improve the performance of machine learning models for cardiovascular risk prediction

Machine learning algorithms have been used in diverse areas among applications, including healthcare. However, to fit an effective and optimal machine learning model, the hyperparameters need to be tuned. This process is commonly referred to as Hyperparameter Optimization and comprises several approaches. We combined three Hyperparameter Optimization techniques (Bayesian Optimization, Particle Swarm Optimization, and Genetic Algorithm) with three classifiers (Random Forest, Support Vector Machine, and XGBoost) to identify the best combination of hyperparameters that maximize model performance. We use the Framingham dataset to test the proposal. For classifier performance, the Support Vector Machine obtained the best result in recall (96.40%) and F-score (93.86%), while XGBoost obtained the best result in precision (96.30%) and specificity (96.36%). In the accuracy metric, both classifiers achieved 95%. Bayesian optimization had the best results in terms of accuracy, precision, specificity, and F-score metrics. Both Particle Swarm Optimization and Genetic Algorithm obtained the best result in the recall metric.

Sentiment Analysis Classification on PLN Mobile Application Reviews using Random Forest Method and TF-IDF Feature Extraction

PT PLN (Persero) has developed the PLN Mobile application to provide electricity services. The large number of users has resulted in various reviews regarding the strengths, weaknesses, and issues of the application. To evaluate the application's quality, sentiment analysis is conducted on user reviews. Review data is obtained through the Google-Play-Scraper API and cleaned through text preprocessing. This study utilizes the TF-IDF feature extraction method and Random Forest for classification. TF-IDF involves weighting each word in a text. This method transforms words into numerical representations, indicating both their frequency and relevance within the document's context. Random Forest is a supervised machine learning algorithm that utilizes ensemble learning which categorizes reviews into positive and negative, This study produced the best model using stemming data and TF-IDF unigram, along with a combination of hyperparameters. The n-estimator was set to 100, max_feature to log2, max_depth to unlimited (none), and entropy criterion, resulting in the highest F1-Score of up to 93.14%.

Performance enhancement of machine learning algorithm for breast cancer diagnosis using hyperparameter optimization

Breast cancer is the most fatal women’s cancer, and accurate diagnosis of this disease in the initial phase is crucial to abate death rates worldwide. The demand for computer-aided disease diagnosis technologies in healthcare is growing significantly to assist physicians in ensuring the effectual treatment of critical diseases. The vital purpose of this study is to analyze and evaluate the classification efficiency of several machine learning algorithms with hyperparameter optimization techniques using grid search and random search to reveal an efficient breast cancer diagnosis approach. Choosing the optimal combination of hyperparameters using hyperparameter optimization for machine learning models has a straight influence on the performance of models. According to the findings of several evaluation studies, the k-nearest neighbor is addressed in this study for effective diagnosis of breast cancer, which got a 100.00% recall value with hyperparameters found utilizing grid search. k-nearest neighbor, logistic regression, and multilayer perceptron obtained 99.42% accuracy after utilizing hyperparameter optimization. All machine learning models showed higher efficiency in breast cancer diagnosis with grid search-based hyperparameter optimization except for XGBoost. Therefore, the evaluation outcomes strongly validate the effectiveness and reliability of the proposed technique for breast cancer diagnosis.

A Real‐time Lithological Identification Method based on SMOTE‐Tomek and ICSA Optimization

AbstractIn petroleum engineering, real‐time lithology identification is very important for reservoir evaluation, drilling decisions and petroleum geological exploration. A lithology identification method while drilling based on machine learning and mud logging data is studied in this paper. This method can effectively utilize downhole parameters collected in realtime during drilling, to identify lithology in real‐time and provide a reference for optimization of drilling parameters. Given the imbalance of lithology samples, the synthetic minority over‐sampling technique (SMOTE) and Tomek link were used to balance the sample number of five lithologies. Meanwhile, this paper introduces Tent map, random opposition‐based learning and dynamic perceived probability to the original crow search algorithm (CSA), and establishes an improved crow search algorithm (ICSA). In this paper, ICSA is used to optimize the hyperparameter combination of random forest (RF), extremely random trees (ET), extreme gradient boosting (XGB), and light gradient boosting machine (LGBM) models. In addition, this study combines the recognition advantages of the four models. The accuracy of lithology identification by the weighted average probability model reaches 0.877. The study of this paper realizes high‐precision real‐time lithology identification method, which can provide lithology reference for the drilling process.

DLSTM-MSF: Distributed LSTM Models for Multimedia Streaming Workload Forecasting Based on Kafka Environment

This paper introduces DLSTM-MSF, a distributed approach designed to address the challenge of demand forecasting in multimedia streaming workloads. DLSTM-MSF leverages the power of multi-LSTM networks, each tailored to predict data demand for a specific type of multimedia streaming workload. The central problem addressed in this research is the accurate prediction of workload demand in a dynamic and diverse multimedia streaming environment. To achieve specialization, the training time series set for each LSTM network comprises examples with targets belonging exclusively to the workload type it is designed to predict. This specialization ensures that each LSTM network becomes proficient at capturing the unique demand patterns associated with its designated workload category. The methodology of the proposed approach is based on building the best forecasting model for each multimedia streaming workload type by exploring various combinations of LSTM hyper-parameters using the grid search method. This enables the proposed approach to effectively capture nonlinear patterns in time series data. Furthermore, the implementation of DLSTM-MSF incorporates Apache Kafka for online demand prediction, utilizing the best-developed model for each workload type. Experimental evaluations of DLSTM-MSF compare the performance of two ensemble-learning LSTM models (Ensemble V1 and Ensemble V2) with a single LSTM model. The results unequivocally highlight the superiority of Ensemble V1, with reductions of 71.85% and 74.88% in RMSE and MAE values, respectively, compared to the single LSTM model. Index Terms— Multimedia streaming, LSTM, Ensemble learning, Forecasting, Workload Demand, Big data.

Re-evaluating GPT-4’s bar exam performance

AbstractPerhaps the most widely touted of GPT-4’s at-launch, zero-shot capabilities has been its reported 90th-percentile performance on the Uniform Bar Exam. This paper begins by investigating the methodological challenges in documenting and verifying the 90th-percentile claim, presenting four sets of findings that indicate that OpenAI’s estimates of GPT-4’s UBE percentile are overinflated. First, although GPT-4’s UBE score nears the 90th percentile when examining approximate conversions from February administrations of the Illinois Bar Exam, these estimates are heavily skewed towards repeat test-takers who failed the July administration and score significantly lower than the general test-taking population. Second, data from a recent July administration of the same exam suggests GPT-4’s overall UBE percentile was below the 69th percentile, and $$\sim$$ ∼ 48th percentile on essays. Third, examining official NCBE data and using several conservative statistical assumptions, GPT-4’s performance against first-time test takers is estimated to be $$\sim$$ ∼ 62nd percentile, including $$\sim$$ ∼ 42nd percentile on essays. Fourth, when examining only those who passed the exam (i.e. licensed or license-pending attorneys), GPT-4’s performance is estimated to drop to $$\sim$$ ∼ 48th percentile overall, and $$\sim$$ ∼ 15th percentile on essays. In addition to investigating the validity of the percentile claim, the paper also investigates the validity of GPT-4’s reported scaled UBE score of 298. The paper successfully replicates the MBE score, but highlights several methodological issues in the grading of the MPT + MEE components of the exam, which call into question the validity of the reported essay score. Finally, the paper investigates the effect of different hyperparameter combinations on GPT-4’s MBE performance, finding no significant effect of adjusting temperature settings, and a significant effect of few-shot chain-of-thought prompting over basic zero-shot prompting. Taken together, these findings carry timely insights for the desirability and feasibility of outsourcing legally relevant tasks to AI models, as well as for the importance for AI developers to implement rigorous and transparent capabilities evaluations to help secure safe and trustworthy AI.

The TPRF: A Novel Soft Sensing Method of Alumina–Silica Ratio in Red Mud Based on TPE and Random Forest Algorithm

The online measurement of the aluminum–silicon ratio of red mud in the dissolution stage of the Bayer alumina production process is difficult to achieve. The offline assay method has a high cost and strong time delay. Soft sensors are an effective and economical method to solve such problems. In this paper, a hybrid model (TPRF model) based on a tree-structured Parzen estimator (TPE) optimized random forest (RF) algorithm is proposed to measure the Al–Si ratio of red mud. The probability distribution of the hyperparameters of the random forest model is estimated by combining the TPE optimization algorithm with the random forest algorithm. According to this probability distribution, the hyperparameters of the random forest algorithm are adjusted in the parameter search space to obtain the best combination of hyperparameters. We established a TPRF soft sensing model based on the optimal combination of hyperparameters. The results show that the best performance of the TPRF model is a mean absolute percentage error (MAPE) of 0.0015, a root-mean-square error (RMSE) of 0.00378, a mean absolute error (MAE) of 0.00162, and a goodness of fit (R2) of 0.9893. The goodness of fit improved by 93.2% compared to the linear model, 39.1% compared to the SVR model, about 21.2% compared to the GRU model, and 5.5% compared to the RF model. This level of performance is demonstrated to be better than traditional soft sensors.

Reinforcement learning for intensive care medicine: actionable clinical insights from novel approaches to reward shaping and off-policy model evaluation

BackgroundReinforcement learning (RL) holds great promise for intensive care medicine given the abundant availability of data and frequent sequential decision-making. But despite the emergence of promising algorithms, RL driven bedside clinical decision support is still far from reality. Major challenges include trust and safety. To help address these issues, we introduce cross off-policy evaluation and policy restriction and show how detailed policy analysis may increase clinical interpretability. As an example, we apply these in the setting of RL to optimise ventilator settings in intubated covid-19 patients.MethodsWith data from the Dutch ICU Data Warehouse and using an exhaustive hyperparameter grid search, we identified an optimal set of Dueling Double-Deep Q Network RL models. The state space comprised ventilator, medication, and clinical data. The action space focused on positive end-expiratory pressure (peep) and fraction of inspired oxygen (FiO2) concentration. We used gas exchange indices as interim rewards, and mortality and state duration as final rewards. We designed a novel evaluation method called cross off-policy evaluation (OPE) to assess the efficacy of models under varying weightings between the interim and terminal reward components. In addition, we implemented policy restriction to prevent potentially hazardous model actions. We introduce delta-Q to compare physician versus policy action quality and in-depth policy inspection using visualisations.ResultsWe created trajectories for 1118 intensive care unit (ICU) admissions and trained 69,120 models using 8 model architectures with 128 hyperparameter combinations. For each model, policy restrictions were applied. In the first evaluation step, 17,182/138,240 policies had good performance, but cross-OPE revealed suboptimal performance for 44% of those by varying the reward function used for evaluation. Clinical policy inspection facilitated assessment of action decisions for individual patients, including identification of action space regions that may benefit most from optimisation.ConclusionCross-OPE can serve as a robust evaluation framework for safe RL model implementation by identifying policies with good generalisability. Policy restriction helps prevent potentially unsafe model recommendations. Finally, the novel delta-Q metric can be used to operationalise RL models in clinical practice. Our findings offer a promising pathway towards application of RL in intensive care medicine and beyond.

Machine learning-based models to predict the conversion of normal blood pressure to hypertension within 5-year follow-up.

Factors contributing to the development of hypertension exhibit significant variations across countries and regions. Our objective was to predict individuals at risk of developing hypertension within a 5-year period in a rural Middle Eastern area. This longitudinal study utilized data from the Fasa Adults Cohort Study (FACS). The study initially included 10,118 participants aged 35-70 years in rural districts of Fasa, Iran, with a follow-up of 3,000 participants after 5 years using random sampling. A total of 160 variables were included in the machine learning (ML) models, and feature scaling and one-hot encoding were employed for data processing. Ten supervised ML algorithms were utilized, namely logistic regression (LR), support vector machine (SVM), random forest (RF), Gaussian naive Bayes (GNB), linear discriminant analysis (LDA), k-nearest neighbors (KNN), gradient boosting machine (GBM), extreme gradient boosting (XGB), cat boost (CAT), and light gradient boosting machine (LGBM). Hyperparameter tuning was performed using various combinations of hyperparameters to identify the optimal model. Synthetic Minority Over-sampling Technology (SMOTE) was used to balance the training data, and feature selection was conducted using SHapley Additive exPlanations (SHAP). Out of 2,288 participants who met the criteria, 251 individuals (10.9%) were diagnosed with new hypertension. The LGBM model (determined to be the optimal model) with the top 30 features achieved an AUC of 0.67, an f1-score of 0.23, and an AUC-PR of 0.26. The top three predictors of hypertension were baseline systolic blood pressure (SBP), gender, and waist-to-hip ratio (WHR), with AUCs of 0.66, 0.58, and 0.63, respectively. Hematuria in urine tests and family history of hypertension ranked fourth and fifth. ML models have the potential to be valuable decision-making tools in evaluating the need for early lifestyle modification or medical intervention in individuals at risk of developing hypertension.

Hyperparameter Optimization of Random Forest Algorithm Technique in Acute Coronary Syndrome Cases

Research using random forest hyperparameter optimization in the case of acute coronary syndrome allows us to obtain a more optimal prediction model, but we can find a gap assumption where the learning carried out by the model still shows symptoms of over-fitting, characterized by a fairly large gap between the training and cross-training processes. validation in the model evaluation process. The research that will be carried out will provide a more optimal prediction model and will not produce symptoms of overfitting of the model using optimization techniques for the hyperparameters in the random forest algorithm. After carrying out various scenarios and testing accuracy, precision scores and various combinations of hyperparameters in the random forest algorithm, it was concluded that the model with the best optimization had a split ratio of 90:10 with an accuracy level of 84.44%, a precision score of 85, 29% and an MSE score of 0.1556 with the results of a combination of random forest optimization hyperparameters using gridCV. The optimization model using random grid cross validation that was built succeeded in reducing the level of over-fitting in the data, decreasing the MSE (mean squared error) from 0.17 and 0.24 to 0.15 for each model.

Predicting the Displacement Variation of Rehabilitated Foundation of Onshore Wind Turbines Using Machine Learning Models

The rehabilitation of wind turbine foundations after damage is increasingly common. However, limited research exists on the deformation of wind turbine foundations after rehabilitation. Artificial intelligence methods can be used to analyze future deformation state and predict post-rehabilitation deformation of foundations. This paper focuses on analyzing the stability of damaged wind turbine foundations after rehabilitation, as well as establishing and evaluating machine learning models. Specifically, Decision Tree (DT), Extreme Gradient Boosting (XGB), Support Vector Regression (SVR), and Long Short-Term Memory Network (LSTM) models are utilized to predict the vertical displacement of the rehabilitated foundation. Hence, the stability of the rehabilitated foundation is discussed in correlation with the measured wind speed, based on the foundation vertical displacement data. During the development of the machine learning model, the most suitable combination of hyperparameters is determined. The prediction performance of the SVR and LSTM models, which exhibit good performance, is compared to further evaluate their effectiveness. Furthermore, the models are analyzed and validated. The results indicate that the vertical displacements of the rehabilitated foundations gradually get close to a state of steady fluctuation over time. The SVR model is identified as the most effective in predicting the vertical displacements of wind turbine foundations after rehabilitation. This study aims to analyze and predict the vertical displacement of wind turbine foundations after rehabilitation based on extensive field monitoring data and powerful machine learning models.