Machine Learning-based Method Research Articles

A Machine Learning-Based Observational Constraint Correction Method for Seasonal Precipitation Prediction

A Machine Learning-Based Observational Constraint Correction Method for Seasonal Precipitation Prediction

Machine Learning-Based Spectral Reconstruction for Equivalence Ratio Measurement in Premixed Air-Methane Flames Using RGB Imaging

ABSTRACT This paper introduced a machine learning-based method for reconstruction spectral information from RGB images for measurements of the equivalence ratio (Φ) of premixed air-methane flames. Digital color cameras capture color and spatial information of hydrocarbon premixed flames in the visible band. The color of the flame is the result of spectral integration of chemiluminescence in the visible wavelength band, and direct use for flame equivalence ratio measurements would result in large errors due to low spectral resolution. The mapping function was used to reconstruct the spectral characteristics of the premixed methane flame for the RGB image, which were built by three types of machine learning methods: support vector regression (SVR), random forest (RF) regression, and Levenberg-Marquardt backpropagation neural network (LM-BPNN), respectively. Finally, LM-BPNN-based reconstructed spectral features are used for the development of Φ measurement model, which measurement error below 0.01, accurately measures the equivalence ratio of premixed methane flames within the range of 0.73 to 1.47. Further analyze the two-dimensional spatial distribution of flame equivalence ratio.

Machine learning-based variable selection for clustered credit risk modeling

AbstractSeveral studies have demonstrated the high prediction accuracy of clustered credit risk modeling. In clustered modeling, borrowers are segmented based on their similarities through cluster analysis, and a separate predictive model is developed for each cluster, resulting in increased predictive accuracy. Unambiguously, its effectiveness depends on the quality of the segmentation, which in turn depends primarily on the choice of variables used in the cluster analysis. However, appropriate variable selection for clustering is a major challenge, particularly for high-dimensional data. In the present study, we propose a machine learning-based variable selection method based on theoretical and regulatory considerations. Formally, the most influential risk drivers from a best-in-class machine learning model are identified using Shapley values and employed as clustering variables. Thus, the information of the explanatory variables crucial for the prediction of the dependent variable is already processed during data segmentation, making each individual predictive model more effective. Through a comparative analysis using two real-world credit default datasets, we show that our proposed approach to clustered modeling leads to the highest prediction accuracy among various clustering models.

A machine learning-based method for predicting the shear behaviors of rock joints

In this study, machine learning prediction models (MLPMs), including artificial neural network (ANN), support vector regression (SVR), K-nearest neighbors (KNN), and random forest (RF) algorithms, were developed to predict the peak shear stress values and shear stress-displacement curves of rock joints. The database used contained 693 records of peak shear stress and 162 original shear stress-displacement curves derived from direct shear tests. The results demonstrated that the MLPMs provided reliable predictions for shear stress, with the mean squared errors (MSEs) between their predicted and measured shear stress varying from 0.003 to 0.069 and the coefficients of determination (R2 values) varying from 0.964 to 0.998. The feature importance values indicate that the joint surface roughness coefficient (JRC) is the most important influential factor in determining the peak shear stress, followed by the joint wall compressive strength (JCS), basic friction angle (φb), and shear surface area (As). Similarly, for the shear stress-displacement curve, the JRC is the dominant factor, followed by As, φb, and JCS. Additional direct shear tests were conducted for model validation. The validation shows that the MLPM predictions demonstrate improved consistency with the experimental results in relation to both the peak shear stress and peak shear displacement.

Machine learning-based non-destructive method for identifying defect causes in OLED displays to enhance productivity

Machine learning-based non-destructive method for identifying defect causes in OLED displays to enhance productivity

Identifying selective PDHK inhibitors using coupled tensor matrix completion and experimental validation

Drug discovery often involves targeting specific members within a family of similar proteins. For example, pyruvate dehydrogenase kinase (PDHK) exists as four isozymes, which exhibit varying expression patterns across multiple tissues. Different PDHK isozymes have been implicated in conditions such as cancer, heart failure, and diabetes, suggesting that targeting them with inhibitors may offer therapeutic benefits. However, simultaneous inhibition of all four PDHK isozymes has the potential to be counterproductive, or poorly tolerated, highlighting a need for isoform-selective PDHK inhibitors. Despite multiple prior reports of PDHK inhibitors, identifying isoform-specific inhibitors for each PDHK isozyme has proven elusive. In this work, we propose a comprehensive framework that combines a machine learning-based prediction method and biochemical testing to screen a library of novel and previously reported compounds, thereby identifying selective inhibitors for PDHK isozymes. Initially, a coupled tensor matrix completion (CTMC) approach is employed to predict compound-target interaction (CTI) and pinpoint target-specific inhibitors. Subsequently, biochemical testing is performed to validate the predicted CTIs. Utilizing this approach, we successfully identified five novel PDHK1-specific inhibitors, underscoring the reliability of this approach as a screening method in the early stages of drug discovery and hit identification.

Prospective and External Validation of an Ensemble Learning Approach to Sensitively Detect Intravenous Fluid Contamination in Basic Metabolic Panels.

Intravenous (IV) fluid contamination within clinical specimens causes an operational burden on the laboratory when detected, and potential patient harm when undetected. Even mild contamination is often sufficient to meaningfully alter results across multiple analytes. A recently reported unsupervised learning approach was more sensitive than routine workflows, but still lacked sensitivity to mild but significant contamination. Here, we leverage ensemble learning to more sensitively detect contaminated results using an approach which is explainable and generalizable across institutions. An ensemble-based machine learning pipeline of general and fluid-specific models was trained on real-world and simulated contamination and internally and externally validated. Benchmarks for performance assessment were derived from in silico simulations, in vitro experiments, and expert review. Fluid-specific regression models estimated contamination severity. SHapley Additive exPlanation (SHAP) values were calculated to explain specimen-level predictions, and algorithmic fairness was evaluated by comparing flag rates across demographic and clinical subgroups. The sensitivities, specificities, and Matthews correlation coefficients were 0.858, 0.993, and 0.747 for the internal validation set, and 1.00, 0.980, and 0.387 for the external set. SHAP values provided plausible explanations for dextrose- and ketoacidosis-related hyperglycemia. Flag rates from the pipeline were higher than the current workflow, with improved detection of contamination events expected to exceed allowable limits for measurement error and reference change values. An accurate, generalizable, and explainable ensemble-based machine learning pipeline was developed and validated for sensitively detecting IV fluid contamination. Implementing this pipeline would help identify errors that are poorly detected by current clinical workflows and a previously described unsupervised machine learning-based method.

Domain knowledge-guided machine learning framework for state of health estimation in Lithium-ion batteries

Accurate estimation of battery state of health is crucial for effective electric vehicle battery management. Here, we propose five health indicators that can be extracted online from real-world electric vehicle operation and develop a machine learning-based method to estimate the battery state of health. The proposed indicators provide physical insights into the energy and power fade of the battery and enable accurate capacity estimation even with partially missing data. Moreover, they can be computed for portions of the charging profile and real-world driving discharging conditions, facilitating real-time battery degradation estimation. The indicators are computed using experimental data from five cells aged under electric vehicle conditions, and a linear regression model is used to estimate the state of health. The results show that models trained with power autocorrelation and energy-based features achieve capacity estimation with maximum absolute percentage error within 1.5% to 2.5%.

Determining plasma dose using equivalent total oxidation potential (ETOP): Concept to practical application via machine learning

Atmospheric pressure nonequilibrium plasma holds significant potential in biomedical applications due to its ability to generate reactive species at low temperatures. However, accurately quantifying and controlling plasma dosage remains challenging. Although equivalent total oxidation potential (ETOP) has been proposed for defining dosage, previous methods required measurement of various reactive oxygen and nitrogen species (RONS) densities, which are impractical in diverse plasma settings. Efficient ETOP prediction across variable conditions is thus essential. To address this, we propose a machine learning-based ETOP modeling method. This study collected RONS density data under various conditions using laser-induced fluorescence and trained an artificial neural network to predict ETOP values based on input parameters like voltage, gas flow rate, oxygen concentration, and humidity. This approach enables efficient ETOP prediction across variable conditions, supporting the standardization and clinical application of plasma medicine.

Characteristics of Machine Learning-based Univariate Time Series Imputation Method

Handling missing values in univariate time series analysis poses a challenge, potentially leading to inaccurate conclusions, especially with frequently occurring consecutive missing values. Machine Learning-based Univariate Time Series Imputation (MLBUI) methods, utilizing Random Forest Regression (RFR) and Support Vector Regression (SVR), aim to address this challenge. Considering factors such as time series patterns, missing data patterns, and volume, this study explores the performance of MLBUI in simulated Autoregressive Integrated Moving Average (ARIMA) datasets. Various missing data scenarios (6%, 10%, and 14%) and model scenarios (Autoregressive (AR) models: AR(1) and AR(2); Moving Average (MA) models: MA(1) and MA(2); Autoregressive Moving Average (ARMA) models: ARMA(1,1) and ARMA(2,2); and Autoregressive Integrated Moving Average (ARIMA) models: ARIMA(1,1,1) and ARIMA(1,2,1)) with different standard deviations (0.5, 1, and 2) were examined. Five comparative methods were also used in this research, including Kalman StructTS, Kalman Auto-ARIMA, Spline Interpolation, Stine Interpolation, and Moving Average. The research findings indicate that MLBUI performs exceptionally well in imputing successive missing values. The results of this study indicate that the performance of MLBUI in imputing consecutive missing values, based on MAPE, yielded values of less than 10% across all scenarios used.

Machine Learning-Based Resource Allocation Algorithm to Mitigate Interference in D2D-Enabled Cellular Networks

Mobile communications have experienced exponential growth both in connectivity and multimedia traffic in recent years. To support this tremendous growth, device-to-device (D2D) communications play a significant role in 5G and beyond 5G networks. However, enabling D2D communications in an underlay, heterogeneous cellular network poses two major challenges. First, interference management between D2D and cellular users directly affects a system’s performance. Second, achieving an acceptable level of link quality for both D2D and cellular networks is necessary. An optimum resource allocation is required to mitigate the interference and improve a system’s performance. In this paper, we provide a solution to interference management with an acceptable quality of services (QoS). To this end, we propose a machine learning-based resource allocation method to maximize throughput and achieve minimum QoS requirements for all active D2D pairs and cellular users. We first solve a resource optimization problem by allocating spectrum resources and controlling power transmission on demand. As resource optimization is an integer nonlinear programming problem, we address this problem by proposing a deep Q-network-based reinforcement learning algorithm (DRL) to optimize the resource allocation issue. The proposed DRL algorithm is trained with a decision-making policy to obtain the best solution in terms of spectrum efficiency, computational time, and throughput. The system performance is validated by simulation. The results show that the proposed method outperforms the existing ones.

Machine learning and FPGA implementation for predicting luminance decay and temperature distribution in micro-LED displays

AbstractA machine learning-based method predicts the luminance decay of micro light emitting diode (micro-LED) displays, utilizing temperature distribution and degradation experiments alongside implementation on field-programmable gate array (FPGA). Micro-LEDs, used in outdoor and indoor billboards, experience degradation due to harsh environmental conditions. To model temperature distribution in indoor advertising displays using minimal data, a temperature model is constructed based on sensor from the panel and thermal images captured by a camera, in relation to the display pattern. The input data is processed using an FPGA, which transmits the sensed temperatures and display patterns to the micro-LED panel. Multilayer Perceptron (MLP), a type of neural network, predicts the temperature distribution over the panel surface, achieving an error of less than 1.1 °C. Separate degradation models forecast luminance decay, factoring in enclosure temperature, input current, and usage time, with distinct models for red, green, and blue LEDs. Exponential curve-fitting and interpolation, following TM-21 standards, ensure long-term accuracy. The luminance decay predictions have an average error below 1.05% (approximately 9 nits). The FPGA implementation minimizes resource consumption while maintaining prediction accuracy, making it suitable for real-time applications. The degradation model accurately predicts performance over tens or even hundreds of thousands of hours, aligning with the exponential decay trends defined by TM-21.

Multiscale Characterization of the Albian-Cenomanian Reservoir System Behavior: A Case Study from the North East Abu Gharadig Basin, North Western Desert, Egypt

Since its discovery in 2010, the NEAG 2 has been one of the most productive oil fields of the Badr El-Din Petroleum Company (BAPETCO) in the northern Western Desert of Egypt. The Albian-Cenomanian reservoir system has a unique performance but suffers from several issues hindering its production. The latest production report in 2023, NEAG-2 Field was producing 1760 bbls of oil with 36500 bbls of water, i.e., 95% water cut. Despite that, the field has reached a 39% recovery factor but the reservoir forecast suggests a much higher recovery factor. Therefore, the NEAG 2 Field requires a comprehensive geological model to depict its reservoir heterogeneities better. We introduce a solid and integrated workflow to investigate the reservoir characters among different scales of geological heterogeneity and offer solutions to overcome some data gaps. After characterizing the reservoir elements by the structural, stratigraphic, petrographic, and petrophysical analyses, a machine learning-based method was applied to overcome the missing whole rock cores in creating a detailed electro-facies log for all field wells. The Neural-Network algorithm required the facies types to be grouped into definitive reservoir qualities to be applied. The resultant electro-facies log had a very good match with the input logs, which validated the facies grouping. This was followed by the porosity-permeability transforms, estimated from mobility data, to create a permeability curve for all field wells, despite the unavailability of core data. The reservoir was categorized into three rock types, each with a specific range of quality, signifying their different flow abilities which were supported by dynamic data. The Lower Bahariya-Kharita in NEAG 2 was ultimately concluded to be a complex heterogeneous reservoir with varying flow abilities and production behaviors. The recovery factor mismatch is due to unrecovered reserves, and a new production strategy should be introduced to reach the ultimate recovery. This integration of geologic and dynamic data is strongly recommended for any reservoir characterization study to avoid oversimplifying the reservoir system and to design the right reservoir development plan.

An Objective Evaluation Method for Image Sharpness Under Different Illumination Imaging Conditions

Blurriness is troublesome in digital images when captured under different illumination imaging conditions. To obtain an accurate blurred image quality assessment (IQA), a machine learning-based objective evaluation method for image sharpness under different illumination imaging conditions is proposed. In this method, the visual saliency, color difference, and gradient information are selected as the image features, and the relevant feature information of these three aspects is extracted from the image as the feature value for the blurred image evaluation under different illumination imaging conditions. Then, a particle swarm optimization-based general regression neural network (PSO-GRNN) is established to train the above extracted feature values, and the final blurred image evaluation result is determined. The proposed method was validated based on three databases, i.e., BID, CID2013, and CLIVE, which contain real blurred images under different illumination imaging conditions. The experimental results showed that the proposed method has good performance in evaluating the quality of images under different imaging conditions.

Deep learning based segmentation of binder and fibers in gas diffusion layers

Gas diffusion layers (GDLs) are vital parts for the performance of proton-exchange membrane fuel cells (PEMFCs). In many cases, they are made of Carbon-Carbon Composite Paper (CCCP), which consists of carbon fibers and a carbonized binder material. The distribution of the fibers and binder in the GDL strongly influences the performance of a PEMFC. Synchrotron scans are a great way to obtain information about the microstructural composition of carbon paper GDLs (Figure 2), but there is one major obstacle. Binder and fibers tend to have the same attenuation and, consequently, the same gray values in the scans. To overcome this, we introduce a machine learning-based method that segments fibers and binder from the local morphology of a CCCP. The training data is generated using FiberGeo, a module in the GeoDict software for fibrous microstructure generation. FiberGeo creates fibers based on stochastic geometry and adds binder using morphological opening and closing operations. We applied the machine learning-based method to four Scans of samples of Toray Carbon Paper with varying amounts of binder in them. The result is the quantification of individual voxels as fiber or binder material that can be used, for example, in performance simulations of property simulations in PEMFCs [1–4]. Here, we focus on the differences in the spatial distribution of the binder both in the through-plane and in-plane directions.

Deciphering the Symptom Spectrum: A Comprehensive Analysis of Migraine Patterns and Types

Complex neurological diseases like migraine migraine affect a large section of the global population, causing health, social, and economic issues. Migraine causes intense, painful headaches that are usually one-sided and pulsing. Auras, nausea, vomiting, and excessive light and sound sensitivity may precede these episodes. Migraine affect millions worldwide and can be intermittent or persistent, impairing function. Diet and stress may induce it, but the cause is unknown. Prevention and symptom treatment drugs and lifestyle changes are used. Debilitating migraines are hard to diagnose due to their varied presentation and subjective symptom reporting. Traditional migraine diagnosis, based on clinical evaluation, typically fails to classify migraine types, requiring more objective and rigorous instruments. This study proposes a machine learning-based migraine categorization method to address this issue. The dataset includes different patient demographics and clinical variables; thus, we use complex algorithms like Random for Forest, XGBoost, and Extra Trees. These algorithms are great for deciphering migraine patterns because they excel at evaluating complex datasets. The research seeks to close this gap to improve migraine classification accuracy, objectivity, and reliability, enabling tailored migraine management and treatment. This neurology study could im- prove migraine diagnosis and treatment with more effective and personalized plans.

The combination of CDX2 expression status and tumor-infiltrating lymphocyte density as a prognostic factor in adjuvant FOLFOX-treated patients with stage III colorectal cancers.

Colorectal carcinomas (CRCs) with caudal-type homeobox 2 (CDX2) loss are recognized to pursue an aggressive behavior but tend to be accompanied by a high density of tumor-infiltrating lymphocytes (TILs). However, little is known about whether there is an interplay between CDX2 loss and TIL density in the survival of patients with CRC. Stage III CRC tissues were assessed for CDX2 loss using immunohistochemistry and analyzed for their densities of CD8 TILs in both intraepithelial (iTILs) and stromal areas using a machine learning-based analytic method. CDX2 loss was significantly associated with a higher density of CD8 TILs in both intraepithelial and stromal areas. Both CDX2 loss and a high CD8 iTIL density were found to be prognostic parameters and showed hazard ratios of 2.314 (1.050-5.100) and 0.378 (0.175-0.817), respectively, for cancer-specific survival. A subset of CRCs with retained CDX2 expression and a high density of CD8 iTILs showed the best clinical outcome (hazard ratio of 0.138 [0.023-0.826]), whereas a subset with CDX2 loss and a high density of CD8 iTILs exhibited the worst clinical outcome (15.781 [3.939-63.230]). Altogether, a high density of CD8 iTILs did not make a difference in the survival of patients with CRC with CDX2 loss. The combination of CDX2 expression and intraepithelial CD8 TIL density was an independent prognostic marker in adjuvant chemotherapy-treated patients with stage III CRC.

Dynamic localization of unwanted events with machine learning-based method

In environmental acoustics measurements it is necessary to separate the contribution of the investigated source from the residual noise, by means of ambient noise measurements. Source separation is a complex task, especially for wind turbine noise, where long-term measurement campaigns are required. Although data analysis procedures that provide for this task are already in use, all of them require a preliminary step for anomalous events removal, i.e. events which are not characteristic of standard residual noise. This step can be complex and very time-expensive depending on the duration of the measurements and the measurement site. Depending on the specific site and source under evaluation, the variety of sounds that can be considered as unwanted for compliance with environmental acoustic regulations is heterogeneous, ranging from road vehicles, trains and planes to human and animal sounds and more. Nowadays, it is possible to extrapolate from an audio source the features that describe its semantics using machine learning. The approach presented in this paper aims to localize in time a specific event within a measurement containing several acoustic events, in order to label it and then proceed to remove the spurious events from the measurement.

Machine learning-Based method for gas leakage source term estimation in highway tunnels

Incidents involving the leakage of flammable and explosive chemicals in highway tunnels can lead to disastrous events such as fires and explosions. The distinctive features of highway tunnels, characterized by their confined and elongated structure, intensify the complexity of managing such accidents, potentially resulting in significant casualties and property damage. Rapid detection and acquisition of leakage source information during a flammable gas leak in a highway tunnel are of utmost importance for effective accident management. This study introduces a predictive model developed using a machine learning approach. The model uses time series data from gas concentration sensors and anemometers installed within the tunnel to predict parameters such as gas leakage velocity, source location, and initiation time. A computational fluid dynamics simulation of gas leakage within the tunnel was constructed, and a dataset was generated using sample generation and preprocessing methods. The results demonstrate that this method exhibits robust predictive performance. Furthermore, a prediction strategy is proposed to enhance the model’s predictive accuracy and resilience to data noise. This model offers valuable insights for improving accident management strategies for gas-related incidents in highway tunnels.

Exploring key factors influencing depressive symptoms among middle-aged and elderly adult population: A machine learning-based method

ObjectiveThis paper aims to investigate the key factors, including demographics, socioeconomics, physical well-being, lifestyle, daily activities and loneliness that can impact depressive symptoms in the middle-aged and elderly population using machine learning techniques. By identifying the most important predictors of depressive symptoms through the analysis, the findings can have important implications for early depression detection and intervention. ParticipantsFor our cross-sectional study, we recruited a total of 976 volunteers, with a specific focus on individuals aged 50 and above. Each participant was requested to provide their demographic, socioeconomic information and undergo several physical health tests. Additionally, they were asked to respond to questionnaires that assessed their mental well-being. Furthermore, participants were requested to maintain an activity log for a continuous 14-day period, starting from the day after they signed up. They had the option to use either a provided mobile application or paper to record their activities. MethodsWe evaluated multiple machine learning models to find the best-performing one. Subsequently, we conducted post-hoc analysis to extract the variable significance from the selected model to gain deeper insights into the factors influencing depression. ResultsLogistic Regression was chosen as it exhibited superior performance across other models, with AUC of 0.807 ± 0.038, accuracy of 0.798 ± 0.048, specificity of 0.795 ± 0.061, sensitivity of 0.819 ± 0.097, NPV of 0.972 ± 0.013 and PPV of 0.359 ± 0.064. The top influential predictors identified in the model included loneliness, health indicator (i.e. frailty, eyesight, functional mobility), time spent on activities (i.e. staying home, doing exercises and visiting friends) and perceived income adequacy. ConclusionThese findings have the potential to identify individuals at risk of depression and prioritize interventions based on the influential factors. The amount of time dedicated to daily activities emerges as a significant indicator of depression risk among middle-aged and elderly individuals, along with loneliness, physical health indicators and perceived income adequacy.