High Predictive Ability Research Articles

Gully erosion prediction method from geoenvironmental data and supervised machine learning techniques

Predictive models are statistical representations that indicate, based on the historical data analysis, the probability of triggering a given phenomenon in the future. In geosciences, such models have been essential to predict the occurrence of adverse phenomena commonly associated with environmental disasters, such as gully erosion. Therefore, this paper presents a method for producing gully erosion predictive models based on geoenvironmental data and machine learning techniques. The method's effectiveness test was produced in a region of approximately 40,000 km² in southeastern Brazil and compared the predictive performance of four models designed with different machine learning algorithms. The results demonstrated that the technique is capable of producing models with high predictive ability, with emphasis on the random forest algorithm, which, in addition to having achieved the highest levels of accuracy, also produced highly realistic maps for the study area.•The method is straightforward and may be applied to predict other geological processes.•The application of the method does not require knowledge of programming language.•The models produced achieved high predictive performance.

A review of machine learning applications in soccer with an emphasis on injury risk

This narrative review paper aimed to discuss the literature on machine learning applications in soccer with an emphasis on injury risk assessment. A secondary aim was to provide practical tips for the health and performance staff in soccer clubs on how machine learning can provide a competitive advantage. Performance analysis is the area with the majority of research so far. Other domains of soccer science and medicine with machine learning use are injury risk assessment, players' workload and wellness monitoring, movement analysis, players' career trajectory, club performance, and match attendance. Regarding injuries, which is a hot topic, machine learning does not seem to have a high predictive ability at the moment (models specificity ranged from 74.2%-97.7%. sensitivity from 15.2%-55.6% with area under the curve of 0.66-0.83). It seems, though, that machine learning can help to identify the early signs of elevated risk for a musculoskeletal injury. Future research should account for musculoskeletal injuries' dynamic nature for machine learning to provide more meaningful results for practitioners in soccer.

Predicting Banana Yield at the Field Scale by Combining Sentinel-2 Time Series Data and Regression Models

Highlights A dataset expansion method based on random sampling could improve the robustness of yield estimation models. CIRE was more suitable for banana yield estimation. XGBoost-based banana yield estimation method showed good prediction ability of banana yield. Abstract. Banana yield prediction at the field level offers significant benefits to growers, packinghouses, crop insurance companies, and researchers. This study explored a remote sensing-based approach for forecasting banana yield at the field scale by using Sentinel-2 (S2) image time series and regression models. First, S2 images of critical phenological periods for bananas were acquired from the Google Earth Engine platform, and these images were treated with cloud and cloud shadow removal. Second, the dataset was expanded by randomly selecting pixels for each field to improve the accuracy of yield prediction. Third, nine vegetation indices (VIs) with high correlation with crop yield were compared and analyzed. Chlorophyll Index Red Edge was selected with a particularly high predictive ability in banana yield prediction. Finally, six regression models, namely, least absolute shrinkage and selection operator (LASSO), support vector regression (SVR), k-nearest neighbors (k-NN), random forest (RF), gradient boosted regression trees (GBRT), and extreme gradient boost (XGBoost), were employed, and their performances were compared. Results showed that the best prediction of banana yield was when 70 pixels were selected for each banana field. Out of nine VIs, comparing different regression models, the XGBoost model emerged as the best learner (the average of R2 for 100 runs in 2019 and 2020 were 0.84 and 0.79, respectively). It was followed by the GBRT model with almost the same performance, which explained 82% and 79% of the banana yield variability for 2019 and 2020, respectively. The LASSO model exhibited the lowest performance of all, but it performed best in terms of stability. The proposed framework applied to satellite image time series can achieve reliable banana yield prediction across years at the field scale. Keywords: Banana yield prediction, Extreme gradient boost, Sentinel-2, Vegetation index.

Association between Anthropometry and Gestational Diabetes Mellitus

Introduction: The predictive value of various anthropometric indicators for identifying benefits or risksfor maternal health outcomes of pregnancy is discussed around the globe. Anthropometric measurements can be a cost effective, efficient method of screening for gestational diabetes mellitus (GDM) especially, in developing countries with low-cost health care settings. Objectives: To determine a relationship between anthropometric measurements and GDM and to assess the importance/suitability of an anthropometric measurement in predicting GDM. Methods: A correlational study with the evaluation of diagnostic test accuracy was conducted among 48 pregnant women at period of amenorrhoea (POA) of 26 - 34 weeks of gestation. The obstetrics ward and the antenatal clinic of Peradeniya teaching hospital was the study setting. Systematic random sampling technique was used for participant selection. Singleton pregnancies with Body Mass Index (BMI) less than 30.0 kg/m2 were included. Women with pre-existing diabetes and medical disorders were excluded. Mid arm circumference (MAC), tricipital skin fold thickness (TSFT), bicipital skin fold thickness (BSFT) were measured according to the National Health and Nutrition Examination Survey (NHANES) anthropometry manual. An interviewer-administered questionnaire was applied to collect the data. Data was evaluated in accordance with the objectives by using SPSS version 25. Results: Mean age of the participants was 29.67 years (SD = 4.76 years). Mean height and weight of the study participants were 154.93 cm and 67.45 kg respectively. Mean BMI value was recorded as 28.13 kgm2. Mean mid arm circumference was 9.43 cm. According to the multivariate analysis done by using logistic regression, calculated TSFT and BSFT values were independently associated with GDM in the population. Successful prediction can be achieved by using the BMI and the body weight (AUC 0.5). 24.8 kg/m2 is taken as the best cut off value to predict GDM (Sn = 79.2; Sp = 29.2). Best cut off value for body weight appears as 60 kg (Sn = 79.2; Sp = 32.3) and the best cut off value for height is 150 cm (Sn = 80.0; Sp = 25.0). When the predictive variables are compared with each other, highest predictive ability was recorded by the body mass index (AUC = 0.632). Predictability of TFT and BSFT appeared significant. 27.0 cm can be considered as the most accurate cut off value of MAC (Sn = 80; Sp = 30). Best cut off values for BSFT and TSFT were 22 mm (Sn = 80; Sp = 60) and 10.5 mm (Sn = 83.3; Sp <span styl

Prediction of methane emissions from fattening cattle using the methane-to-carbon dioxide ratio.

This study aimed to develop a prediction equation for methane (CH4 ) emissions from fattening cattle based on the CH4 /carbon dioxide (CO2 ) ratio and validate the predictive ability of the developed equation. The prediction equation was developed using the CH4 /CO2 ratio combined with oxygen consumption and respiratory quotient estimations that were theoretically calculated from the relation between gas emissions and energy metabolism. To validate the prediction equation, gas measurements in the headboxes were conducted using eight Japanese Black steers. The predictive ability of the developed equation was compared with that of two previously reported equations. As a result, the developed and reported equations had significant (P < 0.01) linear relationships between the observed and predicted CH4 emissions. Notably, only the developed equation had a significant (P < 0.01) linear relationship between the observed and predicted CH4 emissions when expressed per unit of dry matter intake. The results suggest that the developed prediction equation has a higher predictive ability than previously reported equations, particularly in evaluating the efficiency of CH4 emissions. Although further validation is required, the equation developed in this study can be a valuable tool for on-farm estimations of individual CH4 emissions from fattening cattle.

Construction and evaluation of an aging-associated genes-based model for pancreatic adenocarcinoma prognosis and therapies.

Objectives: Pancreatic adenocarcinoma (PAAD) is a highly malignant tumor. Despite extensive research, the precise role of aging-related genes in the initiation, microenvironment regulation, and progression of PAAD remains unclear.Methods: Patients with PAAD were selected from the International Cancer Genome Consortium (ICGC), and The Cancer Genome Atlas (TCGA) cohorts and the cell senescence-associated genes were obtained from CellAge. ConsensusClusterPlus was utilized for cluster identification. The least absolute shrinkage and selection operator (LASSO) Cox regression analysis was performed to construct a prognosis prediction model.Results: We identified three clusters (C1, C2, and C3) based on aging-associated gene profiles. The C1 cluster had a shorter overall survival time, advanced clinical grades, lower immune ESTIMATE score, and tumor immune dysfunction and exclusion (TIDE) score than the C3 subgroup. Moreover, signaling pathways for cell cycle activation were enriched in the C1 cluster. We also identified eight hub genes and constructed a risk model. The high cellular senescence-related signature (CSRS) score subtype exhibited poor prognosis, advanced clinical grades, M2 macrophage infiltration, higher immune checkpoint gene expression, and lower immunotherapeutic benefits.Conclusion: Our risk score model shows high prediction accuracy and survival prediction ability in individual clinical prognosis and pre-immunotherapy evaluation.

Characteristics of m6A-related LncRNAs in breast cancer as prognostic biomarkers and immunotherapy.

N6-methyladenosine (m6A) is a common RNA modification in coding and non-coding RNAs and plays an important role in the occurrence and development of breast cancer (BC). However, the role of m6A-related lncRNAs in breast cancer prognosis is unclear. This study aimed to help verify the biological function of m6A-related lncRNAs in breast cancer prognosis through bio-informatics techniques. First, we screened 18 m6A-related lncRNAs from the TCGA database: AL137847.1, AC137932.2, OTUD6B-AS1, MORF4L2-AS1, AC078846.1, AC012442.1, AL118556.1, AL138955.1, AC009754.1, AC024257.4, AL391095.1, AC024270.3, AC087392.1, LINC02649, AC090948.2, AL158212.1, ITGA6-AS1, AL133243.2 and constructed a risk-prognosis model based on this. Based on the model's median risk score, BC patients were divided into high-risk and low-risk groups. Then, the predictive value of the model was verified by Cox regression, Lasso regression, Kaplan-Meier curve and ROC curve analysis, and biological differences between the two groups were verified by GO enrichment analysis, tumor mutation burden, immune indications and in vitro tests. Importantly, the risk score of this prognostic model is an excellent independent prognostic factor, and m6A regulators are differentially expressed in patients with different risks. In addition, based on patients' different sensitivities to drugs, some drug candidates for different risk populations are screened to provide targets for breast cancer treatment. The difference in immune function between high-risk and low-risk patients also affected the sensitivity to immunotherapy. In the validation of clinical samples, we analyzed the expression of relevant lncRNAs in different risk groups and speculated the possible impact on the prognosis of breast cancer patients. The risk assessment tool built based on the full analysis of these m6A-related genes and m6A-related lncRNA libraries, as well as the m6A-related lncRNAs, has a high prognostic prediction ability, which may provide a supplementary screening method for accurately judging the prognosis of BC and a new perspective for personalized treatment of breast cancer patients.

Concurrent multiscale virtual testing for 2D woven composite structures: A pathway towards composites design and structure optimization

Virtual testing is a powerful tool to characterize the mechanical behavior of materials owing to its excellent predictive ability. However, virtual testing for composite structures remains challenging due to their natural multiscale heterogeneities and expensive computational costs across scales. This paper proposes a FE-SCA concurrent multiscale framework to solve this challenge, where FE and SCA represent the finite element method and the self-consistent clustering analysis, respectively. The mechanical behavior of three woven structures, including open-hole plate, Isoipescu shear, and biaxial tensile specimens, are predicted using the proposed method. The results show that the established FE-SCA concurrent multiscale framework can virtually characterize the damage initiation and evolution for both macroscale woven structures and mesoscale woven representative volume elements. Moreover, it shows significantly higher efficiency compared with the traditional FE 2 framework. Due to the high efficiency and predictive ability, the FE-SCA concurrent multiscale virtual testing has the potential to be a pathway toward composites design and structure optimization.

Morphological and Compositional Features of Chronic Internal Carotid Artery Occlusion in MR Vessel Wall Imaging Predict Successful Endovascular Recanalization.

Background: We sought to determine if the morphological and compositional features of chronic internal carotid artery occlusion (CICAO), as assessed by MR vessel wall imaging (MR-VWI), initially predict successful endovascular recanalization. Methods: Consecutive patients with CICAO scheduled for endovascular recanalization were recruited. MR-VWI was performed within 1 week prior to surgery for evaluating the following features: proximal stump morphology, extent of occlusion, occlusion with collapse, arterial tortuosity, the presence of hyperintense signals (HIS) and calcification in the occluded C1 segment. Multivariate logistic regression was used to identify features associated with technical success and construct a prediction model. Results: Eighty-three patients were recruited, of which fifty-seven (68.7%) were recanalized successfully. The morphological and compositional characteristics of CICAO were associated with successful recanalization, including occlusions limited to C1 and extensive HIS, as well as the absence of extensive calcification, absence of high tortuosity, and absence of artery collapse. The MR CICAO score that comprised the five predictors showed a high predictive ability (area under the curve: 0.888, p < 0.001). Conclusion: the MR-VWI characteristics of CICAO predicted the technical success of endovascular recanalization and may be leveraged for identifying patients with a high probability of successful recanalization.

Измерение массы плода с использованием автоматизированной трехмерной модели фракционного объема конечностей

Objective. To investigate the efficacy of using an automated three-dimensional (3D) fractional limb volume model to calculate estimated fetal weight (EFW) and predict fetal macrosomia. Patients and methods. This prospective observational study included 66 pregnant women who reached 30+0 to 41+3 weeks’ gestation. All patients underwent ultrasonography with fractional thigh volume (TVol) and fractional arm volume (AVol) measurements using 5D Limb Vol software (Samsung Medison W10 ultrasound system). Mean percentage error (MPE) was determined to assess the accuracy of fetal weight estimation. Results. Fractional thigh and arm volumes had a strong correlation with fetal weight (0.834 and 0.617, respectively) and increased linearly with gestational age. When performing a ROC analysis, TVol was found to be a predictor of fetal macrosomia with a sensitivity of 82.35% and specificity of 84.21%, and an experimental model including fetometric parameters (head circumference, abdominal circumference, and femur length) and TVol explained 83% of the variance in birth weight (p = 0.001). In our study, the formulas that included TVol and AVol to calculate EFW were less accurate than the Hadlock formula, with MPE of -1.25 ± 5.75, whereas when using the 3D model with TVol and AVol, MPE was 9.65 ± 7.84 and -9.16 ± 7.63, respectively. However, the experimental model with TVol produced more accurate EFW values compared to the Hadlock formula: the proportion of measurements with an estimated error ≤5% was 94.4% and 100% with an estimated error ≤10% (p &lt; 0.001). Conclusion. Fractional limb volume is an easily reproducible and promising parameter, and its use may be helpful to increase the accuracy of standard fetometry. Fractional thigh volume demonstrated high predictive ability for macrosomia, and its addition to standard fetometric parameters improved the model for fetal weight estimation. Key words: estimated fetal weight, fractional thigh and arm volumes, macrosomia, 5D Limb VolTM

Joint use of population pharmacokinetics and machine learning for optimizing antiepileptic treatment in pediatric population

Purpose: Unpredictable drug efficacy and safety of combined antiepileptic therapy is a major challenge during pharmacotherapy decisions in everyday clinical practice. The aim of this study was to describe the pharmacokinetics of valproic acid (VA), lamotrigine (LTG), and levetiracetam (LEV) in a pediatric population using nonlinear mixed-effect modeling, while machine learning (ML) algorithms were applied to identify any relationships among the plasma levels of the three medications and patients’ characteristics, as well as to develop a predictive model for epileptic seizures. Methods: The study included 71 pediatric patients of both genders, aged 2–18 years, on combined antiepileptic therapy. Population pharmacokinetic (PopPK) models were developed separately for VA, LTG, and LEV. Based on the estimated pharmacokinetic parameters and the patients’ characteristics, three ML approaches were applied (principal component analysis, factor analysis of mixed data, and random forest). PopPK models and ML models were developed, allowing for greater insight into the treatment of children on antiepileptic treatment. Results: Results from the PopPK model showed that the kinetics of LEV, LTG, and VA were best described by a one compartment model with first-order absorption and elimination kinetics. Reliance on random forest model is a compelling vision that shows high prediction ability for all cases. The main factor that can affect antiepileptic activity is antiepileptic drug levels, followed by body weight, while gender is irrelevant. According to our study, children’s age is positively associated with LTG levels, negatively with LEV and without the influence of VA. Conclusion: The application of PopPK and ML models may be useful to improve epilepsy management in vulnerable pediatric population during the period of growth and development.

Linear quantitative structure-ecotoxicity relationship modeling of a series of symmetrical triazine derivatives based on physicochemical parameters

The present study reports the Quantitative Structure-Ecotoxicity Relationship (QSER) analysis of a series of 21 1,3,5-triazine derivatives based on multiple-linear regression (MLR) method. The ecotoxicity data were estimated by using in silico approach and included the following parameters: acute algae toxicity (AAT), acute daphnia toxicity (ADT), Daphnia Magna LC50 48h/EPA (DMepa) and Daphnia Magna LC50 48h/DEMETRA (DMdemetra). The ecotoxicity data were correlated with molecular descriptors selected by using the stepwise selection method. The considered molecular descriptors are lipophilicity descriptors (CrippenLogP, ALogp2), Autocorrelation Descriptor Mass (ATSm1, ATSm2, ATSm3, ATSm4), Autocorrelation Descriptor Charge (ATSc2), minimum E-states for (strong) hydrogen bond acceptors (minHBa), maximum E-states for (strong) hydrogen bond acceptors (maxHBa), second kappa shape index (Kier2), maximum atom-type E-State: ?:N:? (maxaaN), sum of path lengths starting from nitrogens (WTPT-5) and McGowan characteristic volume (McGowan_Volume). The modeling resulted in four statistically valid MLR models. The models were validated by the internal and external validation approaches. The external validation confirmed high predictive ability of the established MLRs.

Networks Data Transfer Classification Based On Neural Networks

Data transmission classification is an important issue in networks communications, since the data classification process has the ultimate impact in organizing and arranging it according to size and area to prepare it for transmission to minimize the transmission bandwidth and enhancing the bit rate. There are several methods and mechanisms for classifying the transmitted data according to the type of data and to the classification efficiency. One of the most recent classification methods is the classification of artificial neural networks (ANN). It is considered one of the most dynamic and up-to-date research in areas of application. ANN is a branch of artificial intelligence (AI). The neural network is trained by backpropagation algorithm. Various combinations of functions and their effect while utilizing ANN as a file, classifier was studied and the validity of these functions for different types of datasets was analyzed. Back propagation neural university (BPNN) supported with Levenberg Marqurdte (LM) activation function might be utilized with as a successful data classification tool with a suitable set of training and learning functions which operates, when the probability is maximum. Whenever the maximum likelihood method was compared with backpropagation neural network method, the BPNN supported with Levenberg Marqurdte (LM) activation function was further accurate than maximum likelihood method. A high predictive ability against stable and well-functioning BPNN is possible. Multilayer feed-forward neural network algorithm is also used for classification. However BPNN supported with Levenberg Marqurdte (LM) activation function proves to be more effective than other classification algorithms.

An immune risk score predicts survival of patients with diffuse large B-cell lymphoma

PurposeProviding accurate prognostic models is necessary for diffuse large B-cell lymphoma, but there are still many uncertainties. So far, none of the models include immune cells. Therefore, an immune risk score was constructed to predict the survival of patients. MethodsCIBERSORTx was chosen to estimate the proportion of 22 human immune cell subsets from public datasets and generate an immune risk score to predict patients' survival in a training cohort using the least absolute shrinkage and selection operator (LASSO) Cox regression model. ResultsThe prognostic model had high predictive ability in the training and validation cohorts. Subjects in the training cohort with high scores had a worse prognosis compared with subjects with low scores. The same result was also found in the three validation cohorts. Multivariable analysis suggested that the immune risk score was an independent prognostic factor. The merged score, including the immune risk score and the international prognostic index (IPI) risk category, had better predictive accuracy. ConclusionsOur immune risk score promises to be a complement to current prognostic models.

Hologram quantitative structure-activity relationship guided in silicon molecular design of phenothiazine dye sensitizers for dye-sensitized solar cells

Hologram quantitative structure–activity relationship (HQSAR) method was successfully introduced into the structure-performance study on the phenothiazine dye sensitizers of dye-sensitized solar cells. A total of 79 phenothiazine derivatives were investigated to build the HQSAR model of power conversion efficiency. The fragment parameters of the established optimal HQSAR model are fragment size: “4 ∼ 7”, hologram length: “353”, and fragment distinction: “bonds, connections, chirality, donor and acceptor atoms”. The predictive performance of the HQSAR model was evaluated with external test set validation and leave-one-out cross-validation. The validation results demonstrate that the developed model possesses high prediction ability and stability. The molecular contribution map of this model was analyzed to reveal the key structures and modification positions for improving the power conversion efficiency, and was used to guide the molecular design of high performance dye sensitizers. The theoretical prediction result of several literature reported phenothiazine compounds indicates that the developed model is an effective and important guidance to the molecular design of dye sensitizers. Therefore, several new phenothiazine derivatives with theoretically improved power conversion efficiency were designed by the proposed HQSAR-based strategy.

Encapsulation of pomegranate peel extract in sucrose matrix by co-crystallization

Many researchers have turned their attention to the polyphenolic content of pomegranate peels. However, polyphenols are unstable compounds and this is the biggest deterrent to their use in the food industry. The aim of this study was the effective encapsulation of pomegranate peel phenolic extract in the matrix of sucrose by co-crystallization utilizing three different experimental designs, the two level factorial design (F), the Central Composite design (CCP), and the Box-Behnken design (ΒΒ), from which the one with the higher predictive ability was selected. The examined parameters were the temperature (125–140 °C) at which the extract is incorporated into the sucrose matrix, the extract soluble solids concentration extract (30–60 °Brix), as well as the dry extract to sucrose ratio (0.1–0.7 g/g). The formed powders were analyzed for their physicochemical properties (moisture content, hygroscopicity, solubility, bulk density, color, antioxidant activity). Both the individual optimum conditions for maximizing efficiency and the composite optimum conditions for simultaneously maximizing efficiency and minimizing moisture content across all three different designs of experiments were determined. Of the studied designs, the CCP was the one with the better predictive ability as it presented the smallest deviations between experimental and predicted efficiency values in both optimum and randomized experiments.

Predictive Modelling for Financial Distress amongst Manufacturing Companies in India

This study develops a model predicting financial distress amongst manufacturing companies in India using logistic regression. 18 financial ratios of 574 companies from 34 industries in manufacturing sector were examined for the period 2005 – 2019 to develop and validate the model. The study can be considered as one of the very few which has examined the financial distress indicators of manufacturing sector in India. EBIT margin, Interest coverage, quick ratio, Cash flow from operations to Sales, Debtors Turnover, Working Capital to Total Assets, Fixed Assets to Total Assets are important determinants of financial health of a business. This study provides useful insights to business managers and lenders to review and monitor financial soundness of business. The findings can also help policy makers to design policies and programs to support distressed industries in India. This study also addresses the urgent need for a country specific model for distress prediction. The model developed shows high predictive ability.&#x0D; Key words: Financial distress, Indian manufacturing sector, distress prediction, logistic regression, distress indicators.

Magnetic resonance enterocolonography for assessing inflammation activity in Crohn’s disease using the CDMI and MEGS indices

AIM: to determine the diagnostic efficiency of magnetic resonance enterocolonography using inflammation indices (CDMI, MEGS) in assessing activity of the inflammatory process in the colon and small intestine in Crohn's disease.PATIENTS AND METHODS: the study included 42 patients with Crohn's disease (age ranged from 19 to 47 years). All patients underwent magnetic resonance enterocolonography (MRE) with intravenous contrast and diffusion-weighted images. According to the results of MRE, were determined MR indices of inflammation activity in the small and large intestine - CDMI and MEGS. The MR inflammatory activity indices CDMI and MEGS were compared with the endoscopic inflammatory activity index SES-CD.RESULTS: the MR inflammation activity index CDMI did not show a statistically significant diagnostic efficiency (p&gt;0.05). The MR inflammation activity index MEGS showed a statistically significant diagnostic efficiency (p&lt;0.0001). According to the nomogram, the MR inflammation activity index MEGS demonstrated a high predictive ability to determine the true activity of the inflammatory process in the small and large intestine. Correlation demonstrated direct strong relationship between the quantitative values of the MEGS inflammation indices and the endoscopic inflammation index (SES-CD) (r=0.843, p&lt;0.0001).CONCLUSION: magnetic resonance enterocolonography using the MEGS activity index has a high diagnostic efficiency in assessing the activity of the inflammatory process in the small and large intestine in Crohn's disease. The MR inflammation activity index CDMI did not show statistically significant diagnostic efficacy. The results of the study should be considered preliminary and require further study of the possibilities of these indices of inflammation activity in a larger sample of patients.

Modelling the impact of disease outbreaks on the international crude oil supply chain using Random Forest regression

PurposeThe crude oil supply chain (COSC) is one of the most complex and largest supply chains in the world. It is easily vulnerable to extreme events. Recently, the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) (often known as COVID-19) pandemic created a massive imbalance between supply and demand which caused significant price fluctuations. The purpose of this study is to explore the influential factors affecting the international COSC in terms of consumption, production and price. Furthermore, it develops a model to predict the international crude oil price during disease outbreaks using Random Forest (RF) regression.Design/methodology/approachThis study uses both qualitative and quantitative approaches. A qualitative study is conducted using a literature review to explore the influential factors on COSC. All the data are extracted from Web sources. In addition to COVID-19, four other diseases are considered to optimize the accuracy of predictive results. A principal component analysis is deployed to reduce the number of variables. A forecasting model is developed using RF regression.FindingsThe findings of the qualitative analysis characterize the factors that influence international COSC. The findings of quantitative analysis emphasize that production and consumption have a higher contribution to the variance of the data set. Also, this study found that the impact caused to crude oil price varies with the region. Most importantly, the model introduced using the RF technique provides a high predictive ability in short horizons such as infectious diseases. This study delivers future directions and insights to researchers and practitioners to expand the study further.Originality/valueThis is one of the few available pieces of research which uses the RF method in the context of crude oil price forecasting. Additionally, this study examines international COSC in the events of emergencies, specifically disease outbreaks using machine learning techniques.

Local interpretation of nonlinear regression model with k-nearest neighbors

With respect to molecular, material, and process designs, it is important to construct nonlinear regression models with high predictive ability between the features, that is, x, and the properties and activities, that is, y. The interpretations of such constructed models can help elucidate the mechanism by which x affects y. In this study, a stable and effective method for the local interpretation of regression models is proposed by improving the local interpretable model-agnostic explanations (LIME) and Shapley additive explanations (SHAP) techniques, which have been developed as methods for interpreting nonlinear regression models. By calculating the local contribution of x to y based on LIME and SHAP for the k-nearest neighbors of a target sample, a stable interpretation is possible even for models that overfit the training data. Furthermore, a method for calculating the local contribution of x to y based on simulations using a nonlinear regression model is proposed. Finally, using actual datasets consisting of the characteristics of various compounds, it is confirmed that the proposed method can locally and accurately interpret nonlinear models.Tweetable abstractA method for calculating the local contribution of x to y based on simulations using a nonlinear regression model is proposed, and could accurately interpret nonlinear models for actual datasets.