Variable Selection Process Research Articles

Detection and severity of COVID-19 cases based on patient symptoms Using decision trees

At the end of 2019, a type of pneumonia of unknown origin was detected in Wuhan, China. It was later determined that the illness was caused by the SARS-CoV-2 virus, and in 2020, the World Health Organization designated this disease as COVID-19. Various efforts have been made to enable timely detection of COVID-19 within the field of computational systems. This study proposes detecting COVID-19 based on the symptoms experienced by patients, utilizing data provided by the National Epidemiological Surveillance System (SINAVE) of Mexico City, from which 403,185 records were used. In situations where positive cases of COVID-19 are detected, it is predicted whether it will be a serious case in which the patient needs to be intubated or admitted to the Intensive Care Unit. For this, classification and regression decision trees (CART) are used. Different parameters were considered to define the CART model, and the stepwise variable selection process was also used to determine the significant variables that offer the best results, obtaining an accuracy of 87.04%. This study shows progress in the detection of COVID-19 using only the symptoms presented by the patients.

P-802. The Predictive Accuracy of Clinical Covariates in Differentiating Staphylococcus aureus Bacteremia with and without Deep-Seated Foci of Infection via Application of Different Variable Selection Methodologies

Abstract Background Staphylococcus aureus bacteremia (SAB) is a common bloodstream infection with a mortality rate of 20-30%. SAB is frequently associated with deep-seated foci of infection (DSFI) for which IDSA/ESCMID draft guidelines propose using clinical features early in evaluation to risk stratify for DSFI. However, literature for using combinations of clinical features to identify DSFI is limited, and the best process of variable selection is not defined. We explored variable selection methods to identify a combination of clinical features with the best predictive accuracy for identifying SAB with DSFI. Methods 242 adults admitted to a 500-bed teaching hospital in Portland, Oregon with incident SAB between January 1, 2016 and December 31, 2017 were included. Patients who died, transitioned to comfort or discharged before medically advised within 7 days, or without an infectious disease consultation were excluded. Clinical information was abstracted from electronic medical records. We performed four variable selection methods: best subset, forward and backward stepwise regression, and the LASSO. Variables were selected using 5-fold cross-validation. We estimated the area under the receiver operating curve (AUC) and fit a multivariable logistic regression model to evaluate the expected performance of the predictor combinations. Results The cohort is described in Table 1 and 2. SAB with one or more DSFI was diagnosed in 65% (n=158). Predictors selected by the four methods are described in Table 3. The combination of bacteremia duration, hospital acquired SAB (haSAB), central venous catheter (CVC), skin and soft tissue infection (SSTI), and any cardiac device had the highest predictive accuracy with an AUC of 82%. The odds of DSFI is estimated to increase by 32% (95% CI; 15% to 53%) for every 1 day increase in bacteremia after adjusting for haSAB, CVC, SSTI and any cardiac device (Table 4). Conclusion Duration of bacteremia, haSAB, and CVC were included regardless of the variable selection method. Our model shows that combinations of clinical features are more predictive of DSFI than any single feature. Larger studies in different cohorts are needed to verify the predictive accuracy of these variables used in combinations to risk stratify for SAB with DSFI. Disclosures All Authors: No reported disclosures

Lower ungulate population density in rainforests under anthropogenic influences

Abstract The main threats to forest ungulates are land cover change, poaching, invasive species and other anthropogenic factors. Tropical rainforest deer, especially those in the South American Atlantic Forest, are extremely elusive, resulting in limited data on their density and the factors influencing it. In this study, our objective was to estimate the density of Atlantic Forest deer and to identify its most significant drivers. We conducted density estimates using faecal standing crop methods with scat detection dogs across the whole Atlantic Forest. In addition, we investigated the influences on density of a locally collected environmental index (AII) incorporating anthropogenic impact variables, landscape metrics (habitat cover, matrix heterogeneity and number of habitat patches), altitude, slope, canopy height and socioeconomic indexes (human development index, purchasing power parity, income inequality and gender development index) as well as protected area ranger density. After a variable selection process to avoid autocorrelation, we tested the relationship between the selected variables and forest deer population density using multiple linear regression models. Density varied from 0.14 to 18.17 individuals/km2, with P. nemorivagus showing the highest values. The average and median values were 3.42 and 1.47 individuals/km2, respectively. Only the model where density was a function of AII was significant, with satisfactory residuals and the lowest AIC value. Synthesis and applications. Our results highlight how threats such as poaching, predation by domestic dogs, disease transmission from domestic ungulates and competition, predation or disturbance by wild boar may affect forest deer density and potentially lead to species extinction.

A novel importance scores based variable selection approach and validation using a MIR and NIR dataset.

Variable selection is important in spectral analysis for improving interpretation quality and accuracy. This study introduces a novel variable selection process, named "VMHBSC", which consists of six steps, with each letter representing one step. To demonstrate its process and advantages, two datasets were employed, a mid-infrared spectral (MIR) dataset (234×7468, sample number×variables) of Chenpi samples (a traditional Chinese medicinal material derived from the dried peel of mature tangerines) and a near-infrared spectral (NIR) dataset (16000×256) for modeling competition. In the MIR dataset, VMHBSC selected 3 important variables from all 7468 variables, and models established using Decision Trees (DT), Gradient Boosting Decision Tree (GBDT), and Extreme Gradient Boosting (XGBoost) achieved higher accuracy compared to models using other variable selection methods. For the NIR dataset, VMHBSC selected 24 important variables from all 256 variables. Based on these 24 common variables, three hybrid models (VMHBSC-DT, VMHBSC-GBDT and VMHBSC-XGBoost) were also established and shown stable performance. These findings indicate the effectiveness of the VMHBSC process in enhancing model performance and robustness.

Complex multivariate model predictions for coral diversity with climatic change

AbstractModels of the future of coral reefs are potentially sensitive to theoretical assumptions, variable selectivity, interactions, and scales. A number of these aspects were evaluated using boosted regression tree models of numbers of coral taxa trained on ~1000 field surveys and 35 spatially complete influential environmental proxies at moderate scales (~6.25 km2). Models explored influences of climate change, water quality, direct human‐resource extraction, and variable selection processes. We examined the predictions for numbers of coral taxa using all variables and compared them to models based on variables commonly used to predict climate change and human influences (eight and nine variables). Results indicated individual temperature variables alone had lower predictive ability (R2 < 2%–7%) compared to human influence variables (6%–18%) but overall climate had a higher training–testing fit (70%) than the human influence (63%) model. The full variable model had the highest fit to the full data (27 variables; R2 = 85%) and indicated the strongly interactive and complex role of environmental and human influence variables when making moderate‐scale biodiversity predictions. Projecting changes using Coupled Model Intercomparison Project (CMIP) 2050 Representative Concentration Pathways (RCP2.6 and 8.5) water temperature predictions indicated high local variability and fewer negative effects than predictions made by coarse scale threshold and niche models. The persistence of coral reefs over periods of rapid climate change is likely to be caused by smaller scale variability that is poorly simulated with coarse scale modeled predictions.

A Bayesian approach for variable selection in mixture of logistic regressions with Pólya-Gamma data augmentation

We present Bayesian methods for estimating and selecting variables in a mixture of logistic regression models. A common issue with the logistic model is its intractable likelihood, which prevents us from applying simpler Bayesian algorithms, such as Gibbs sampling, for estimating and selecting the model since there is no conjugacy for the regression coefficients. We propose to solve this problem by applying the data augmentation approach with Pólya-Gamma random variables to the logistic regression mixture model. For selecting covariates in this model, we investigate the performance of two prior distributions for the regression coefficients. A Gibbs sampling algorithm is then applied to perform variable selection and fit the model. The conjugacy obtained for the distribution of the regression coefficients allows us to analytically calculate the marginal likelihood and gain computational efficiency in the variable selection process. The methodologies are applied to both synthetic and real data.

Abstract 4141497: Left ventricular diastolic function by echocardiography is related to extracellular volume and is an independent predictor of mortality in patients with aortic regurgitation

Background: Cardiac catheterization along with cardiac biopsy studies have shown the presence of abnormally increased left ventricular (LV) chamber stiffness in patients with severe aortic regurgitation (AR) and its significant correlation with interstitial fibrosis. LV diastolic dysfunction (DD) is associated with increased mortality in several patient populations. However, there is a paucity of data on the association of DD with mortality in patients with AR. Hypotheses: 1 - LV DD is related to extracellular volume (ECV), imaging marker of interstitial fibrosis, by cardiac magnetic resonance imaging (CMR) 2 - LV DD by echocardiography is associated with mortality in patients with AR Methods: We studied patients with moderate or greater AR who underwent echocardiographic and CMR imaging within 12 months. We excluded patients with concomitant valve disease, prior valve intervention, cardiomyopathy deemed unrelated to AR, congenital heart disease and terminal conditions. The primary outcome was all-cause death. CMR was used to obtain LV volumes, EF and mass, AR severity, scar extent by delayed hyperenhancement, and ECV. DD was assessed by comprehensive echocardiography. Results: The study included 323 patients. Mean age was 61.9 ± 16.0 years and 22% were female. Median follow-up was 3.8 years, during which 65 patients died and 172 had aortic valve replacement. Median LV EF was 56 (47-65) %, AR volume on CMR was 40 (30-61) mL, and AR fraction was 40 (33-47) %. ECV progressively increased with higher grades of LV DD (Figure 1). Table 1 shows the univariate associations with mortality in the study. On multivariable analysis based on the Lasso variable selection process, Euroscore, Charlson comorbidity index, DD grade, and RAP were associated with total mortality. Conclusion: ECV by CMR progressively increases with higher grades of DD by echocardiography, and DD is independently associated with mortality in patients with AR.

A Forest Fire Prediction Model Based on Meteorological Factors and the Multi-Model Ensemble Method

More than half of South Korea’s land area is covered by forests, which significantly increases the potential for extensive damage in the event of a forest fire. The majority of forest fires in South Korea are caused by humans. Over the past decade, more than half of these types of fires occurred during the spring season. Although human activities are the primary cause of forest fires, the fact that they are concentrated in the spring underscores the strong association between forest fires and meteorological factors. When meteorological conditions favor the occurrence of forest fires, certain triggering factors can lead to their ignition more easily. The purpose of this study is to analyze the meteorological factors influencing forest fires and to develop a machine learning-based prediction model for forest fire occurrence, focusing on meteorological data. The study focuses on four regions within Gangwon province in South Korea, which have experienced substantial damage from forest fires. To construct the model, historical meteorological data were collected, surrogate variables were calculated, and a variable selection process was applied to identify relevant meteorological factors. Five machine learning models were then used to predict forest fire occurrence and ensemble techniques were employed to enhance the model’s performance. The performance of the developed forest fire prediction model was evaluated using evaluation metrics. The results indicate that the ensemble model outperformed the individual models, with a higher F1-score and a notable reduction in false positives compared to the individual models. This suggests that the model developed in this study, when combined with meteorological forecast data, can potentially predict forest fire occurrence and provide insights into the expected severity of fires. This information could support decision-making for forest fire management, aiding in the development of more effective fire response plans.

StableMate: a statistical method to select stable predictors in omics data.

Identifying statistical associations between biological variables is crucial to understanding molecular mechanisms. Most association studies are based on correlation or linear regression analyses, but the identified associations often lack reproducibility and interpretability due to the complexity and variability of omics datasets, making it difficult to translate associations into meaningful biological hypotheses. We developed StableMate, a regression framework, to address these challenges through a process of variable selection across heterogeneous datasets. Given datasets from different environments, such as experimental batches, StableMate selects environment-agnostic (stable) and environment-specific predictors in predicting the response of interest. Stable predictors represent robust functional dependencies with the response, and can be used to build regression models that make generalizable predictions in unseen environments. We applied StableMate to (i) RNA sequencing data of breast cancer to discover genes that consistently predict estrogen receptor expression across disease status; (ii) metagenomics data to identify microbial signatures that show persistent association with colon cancer across study cohorts; and (iii) single-cell RNA sequencing data of glioblastoma to discern signature genes associated with thedevelopment of pro-tumour microglia regardless of cell location. Our case studies demonstrate that StableMate is adaptable to regression and classification analyses and achieves comprehensive characterization of biological systems for different omics data types.

Metabolomic Changes Associated with AGXT2 Genotype Variants and Stone Formation in a Colony of Cats.

The objective of this study was to assess serum chemistries and metabolomic parameters in cats with genetic variants of the alanine-glyoxylate aminotransferase 2 (AGXT2) gene to determine abnormalities associated with urolith formation and better understand effective approaches for the treatment of cats with uroliths. AGXT2 genotypes of 445 cats in the colony at Hill's Pet Nutrition, Inc. (Topeka, KS, USA) were assessed in a genome-wide association study. Additionally, the serum chemistries and metabolic profiles of each cat were determined, along with their lifetime history of stone incidence. Factor analysis was used as a data-reduction method for metabolites in order to perform statistical hypothesis testing and to select significant metabolites from the more than 600 serum metabolites identified. Of the 82 cats forming stones in the colony (18.4%), the majority were calcium oxalate. Results showed that approximately one third of the cats with the AA variant of the AGXT2 gene have stones, that chronic kidney disease (CKD) is more common in cats with stones, and that having stones results in a shorter lifespan. A discriminant variable selection process was performed to determine the complete blood count, serum biochemistries, and serum metabolomic factors that best discriminated among the three genotypes (AA, AG, GG) and between cats forming stones and non-stone formers. Several of the highly ranked discriminating factors included metabolites related to decreased aminotransferase activity in cats with the AA variant of the AGXT2 gene. Another factor that ranked highly for discriminating between stone formers and non-stone formers contained lipid metabolites, consisting of multiple sphingomyelin species and cholesterol. These findings support the results of feeding studies in cats, whereby CKD cats fed food supplemented with betaine and prebiotics have experienced an increase in total body mass, reduced uremic toxins, and altered sphingomyelin concentrations.

A multi-layer nesting and integration approach for predicting groundwater levels in agriculturally intensive areas using data-driven models

Agricultural water demand, groundwater extraction, surface water delivery and climate exhibit complex nonlinear relationships with groundwater storage in agricultural regions. As alternatives to computationally intensive physical models, data-driven machine learning methods are frequently employed as surrogates to capture these complex relationships, owing to their high computational efficiency. Inevitably, reliance on a single machine learning model may lead to underestimation of prediction uncertainty and potentially result in reduced accuracy. This paper presents a multi-layer nesting and integration approach for predicting groundwater levels in agriculturally intensive areas using data-driven models. The key contributions of the research are threefold: 1) the development of a comprehensive input variable selection process considering the lag effects and driving mechanisms of groundwater level variations; 2) the implementation of an optimization-based hyperparameter tuning method to enhance the performance of individual machine learning models; and 3) the establishment of a multi-model integration framework based on a multi-layer nesting technique. This approach combines the outputs of multiple machine learning models to consolidate predictions and expand the hypothesis space. The effectiveness of the proposed framework is demonstrated through case studies in the Huaihe River Basin, a major agricultural region in China. The results show that the multi-layer nesting and integration approach outperforms the use of individual machine learning models, providing more accurate and reliable groundwater level predictions. This framework offers valuable insights for decision-makers and water resource managers, supporting sustainable groundwater management and addressing the challenges faced in agriculturally intensive areas.

Explainable Machine Learning Models for Predicting Recurrence in Differentiated Thyroid Cancer

Aim: Differentiated thyroid cancer (DTC) is a common type of cancer that originates in the thyroid gland. This study aimed to predict the recurrence of differentiated thyroid carcinoma, in patient with well-DTC, using explainable machine learning (XAI) models. Material and Methods: The study utilized a dataset from the UCI Machine Learning Repository, which included 383 patients and 13 candidate predictors. After a variable selection process using distance correlation, only four predictors (Response, Risk, T, and N) were retained for model building. Two XAI models, Fast Interpretable Greedy-Tree Sums (FIGS) and Explainable Boosting Machines (EBM), were employed. Results: The EBM model slightly outperformed the FIGS model in terms of accuracy. The study found that the most influential predictors of Well-DTC recurrence were the response to DTC treatment, risk status according to the American Thyroid Association classification, tumor size (T), and lymph node metastasis (N). Conclusion: In conclusion, this study successfully identified key risk factors for DTC recurrence using XAI models, providing interpretable insights for clinical decision-making and potential for personalized treatment strategies.

Investigation of Precepting-Related Burnout in Dietetics Preceptors

BackgroundDietetics preceptors are at risk of abandoning their role because of potential burnout. Identifying factors associated with greater and lesser burnout can help key stakeholders identify and implement strategies to prevent or mitigate burnout in dietetics preceptors. ObjectiveThe purpose of this study was to identify factors associated with the burnout preceptors attribute specifically to the work of precepting. DesignThis study used a cross-sectional study design. A survey instrument including burnout measures, precepting measures, and items querying personal and work characteristics was distributed via e-mail to a random sample of 10 000 credentialed nutrition and dietetics practitioners. Participants settingRespondents who lived in the United States, were currently practicing, and had precepted during the past year were included in the sample (n = 310). The survey was conducted in October and November of 2022. Main outcome measuresPrecepting-related burnout was measured using scores on the Copenhagen Burnout Inventory; scores ranged from 0 to 100 and higher numbers indicated greater burnout. Statistical analyses performedFactors associated with precepting-related burnout were identified using an analysis of covariance. Pairwise comparisons with Tukey adjustments were done to determine differences across categories for the categorical variable in the model (ie, percent of interns requiring additional coaching). A stepwise variable selection process was performed to determine the best analytic models. ResultsFactors associated with lower precepting-related burnout in dietetics preceptors included higher Commitment to the Preceptor Role scores, lower percentages of interns requiring additional coaching, feeling appreciated by the dietetic interns, and higher Preceptors’ Perceptions of Support scores. A factor associated with higher precepting-related burnout was precepting a greater number of dietetic interns in the past year. For every 1-point change in each variable, the degree of change in precepting-related burnout was –7.9 for Commitment to the Preceptor Role, –3.2 for feeling appreciated by dietetic interns, –4.5 for Preceptors’ Perceptions of Support, and +1.2 for number of dietetic interns per year. The mean ± SE difference in precepting-related burnout scores between preceptors who reported < 10% of dietetic interns requiring additional coaching and those who reported > 50% of dietetic interns requiring additional coaching was –13.7 ± 3.7. ConclusionsThe factors found to be associated with precepting-related burnout scores in dietetics preceptors are potentially modifiable, suggesting this burnout might be prevented or mitigated to some degree.

Weighting the United States All of Us Research Program data to known population estimates using raking

BackgroundThe All of Us Research Program aims to collect longitudinal health-related data from a million individuals in the United States. An inherent challenge of a non-probability sampling strategy through voluntary participation in All of Us is that findings may not be nationally representative for addressing health and health care at the population level. We generated survey weights for the All of Us data that can be used to address the challenge. Research designWe developed raked weights using demographic, health, and socioeconomic variables available in both the 2020 National Health Interview Survey (NHIS) and All of Us. We then compared the unweighted and weighted prevalence of a set of health-related variables (health behaviors, health conditions, and health insurance coverage) estimated from All of Us data with the weighted prevalence estimates obtained from NHIS data. SubjectsThe sample included 100,391 All of Us participants 18 years of age and older with complete data collected between May 2017 and January 2022 across the United States. ResultsFinal variables in the raking procedure included age, sex, race/ethnicity, region of residence, annual household income, and home ownership. The mean percentage difference between known proportions obtained from the NHIS and All of Us was reduced by 18.89% for health-related variables after applying the raked weights. ConclusionsRaking improved the comparability of prevalence estimates obtained from All of Us to known national prevalence estimates. Refining the process of variable selection for raking may further improve the comparability between All of Us and nationally representative data.

Opportunities offered by latent-based multiblock strategies to integrate biomarkers of chemical exposure and biomarkers of effect in environmental health studies

Modern environmental epidemiology benefits from a new generation of technologies that enable comprehensive profiling of biomarkers, including environmental chemical exposure and omic datasets. The integration and analysis of large and structured datasets to identify functional associations is constrained by computational challenges that cannot be overcome using conventional regression methods. Some extensions of Partial Least Squares (PLS) regression have been developed to efficently integrate multiple datasets, including Multiblock PLS (MB-PLS) and Sequential and Orthogonalized PLS; however, these approaches remain seldom applied in environmental epidemiology. To address that research gap, this study aimed to assess and compare the applicability of PLS-based multiblock models in an observational case study, where biomarkers of exposure to environmental chemicals and endogenous biomarkers of effect were simultaneously integrated to highlight biological links related to a health outcome. The methods were compared with and without sparsity coupling two metrics to support the variable selection: Variable Importance in Projection (VIP) and Selectivity Ratio (SR). The framework was applied to a case-study dataset mimicking the structure of 36 environmental exposure biomarkers (E-block), 61 inflammation biomarkers (M-block), and their relationships with the gestational age at delivery of 161 mother-infant pairs. The results showed an overall consistency in the selected variables across models, although some specific selection patterns were identified. The block-scaled concatenation-based approaches (e.g. MB-PLS) tended to select more variables from the E-block, while these methods were unable to identify certain variables in the M-block. Overall, the number of variables selected using the SR criterion was higher than using the VIP criterion, with lower predictive performances. The multiblock models coupled to VIP, appeared to be the methods of choice for identifying relevant variables with similar statistical performances. Overall, the use of multiblock PLS-based methods appears to be a good strategy to efficiently support the variable selection process in modern environmental epidemiology.

Improved nonparametric survival prediction using CoxPH, Random Survival Forest & DeepHit Neural Network

In recent times, time-to-event data such as time to failure or death is routinely collected alongside high-throughput covariates. These high-dimensional bioinformatics data often challenge classical survival models, which are either infeasible to fit or produce low prediction accuracy due to overfitting. To address this issue, the focus has shifted towards introducing a novel approaches for feature selection and survival prediction. In this article, we propose a new hybrid feature selection approach that handles high-dimensional bioinformatics datasets for improved survival prediction. This study explores the efficacy of four distinct variable selection techniques: LASSO, RSF-vs, SCAD, and CoxBoost, in the context of non-parametric biomedical survival prediction. Leveraging these methods, we conducted comprehensive variable selection processes. Subsequently, survival analysis models—specifically CoxPH, RSF, and DeepHit NN—were employed to construct predictive models based on the selected variables. Furthermore, we introduce a novel approach wherein only variables consistently selected by a majority of the aforementioned feature selection techniques are considered. This innovative strategy, referred to as the proposed method, aims to enhance the reliability and robustness of variable selection, subsequently improving the predictive performance of the survival analysis models. To evaluate the effectiveness of the proposed method, we compare the performance of the proposed approach with the existing LASSO, RSF-vs, SCAD, and CoxBoost techniques using various performance metrics including integrated brier score (IBS), concordance index (C-Index) and integrated absolute error (IAE) for numerous high-dimensional survival datasets. The real data applications reveal that the proposed method outperforms the competing methods in terms of survival prediction accuracy.

Time varying effects in survival analysis: a novel data-driven method for drift identification and variable selection

In this paper we address the problem of survival models when high-dimensional panel data are available. We discuss two related issues: The first one concerns the issue of variable selection and the second one deals with the stability over time of such a selection, since presence of time dimension in survival data requires explicit treatment of evolving socio-economic context. We show how graphical models can serve two purposes. First they serve as the input for a first algorithm to to assess the temporal stability of the data: Secondly, allow the deployment of a second algorithm which partially automates the process of variable selection, while retaining the option to incorporate domain expertise in the process of empirical model-building. To put our proposed methodology to the test, we utilize a dataset comprising Italian firms funded in 2009 and we study the survival of these entities over the period of 10 years. In addition to revealing significant volatility in the set of variables explaining firm exit over the years, our novel methodology enables us to offer a more nuanced perspective than the conventional one regarding the critical roles played by traditional variables such as industrial sector, geographical location, and innovativeness in firm survival.

Prediction of physicochemical properties of cape gooseberry (Physalis peruviana L.) using near infrared hyperspectral imaging (NIR-HSI)

Cape gooseberries possess high nutritional value, characterized by elevated levels of vitamin C. This study proposed the application of near infrared hyperspectral imaging (NIR-HSI) coupled with chemometrics (Partial Least Square Regression (PLSR) and Support Vector Machine Regression (SVMR)) to predict vitamin C content (mg/mL), firmness (N), soluble solids content (SSC) (°Brix), and titratable acidity (TA) ( %). Vitamin C content was the property with best prediction performance (RMSEP = 4.15 (mg/mL) and RPD = 2.61.) SVMR for predicting firmness achieved an RPD of 2.31, while for predicting SSC and titratable acidity, RMSEP values were <1 %. The bottom/calyx of berries proved to be the best region for image acquisition. Overall, full-spectral-based PLSR and SVMR exhibited comparable performance regarding RPD values. However, following a meticulous variable selection process, SVMR demonstrated superior adaptability to dimensionality reduction, increasing +17 % RPD values. Finally, we demonstrate that NIR-HSI could predict the physicochemical composition of Cape gooseberries.

Prediction enhancement for surface water sodium adsorption ratio using limited inputs: Implementation of hybridized stacked ensemble model with feature selection algorithm

The Sodium Adsorption Ratio (SAR) is a widely used variable in water quality research, particularly in agriculture and environmental studies. In many cases, the key variables required for SAR calculation, namely Na+, Mg+2, and Ca+2, are not available. Consequently, the potential to calculate SAR using a limited number of water quality variables becomes critically important. The study implemented the Multilayer Perceptron Neural Network (MLPNN), Support Vector Regression (SVR), and K-Nearest Neighbors (KNN) models at level-0 for prediction purposes, along with the Boruta model for variable selection. A stacked ensemble learning model at level-1 enhanced the prediction accuracy. The discharge and water quality dataset from the Zarrin-Gol River in northern Iran was utilized to implement the modeling procedure. Results obtained from the variable selection process using the Boruta model revealed that using a limited number of water quality variables can effectively predict SAR even without the principal variables. Further investigation of the input combinations for the level-0 models demonstrated that, for the MLPNN, KNN, and SVR models, 4, 3, and 1 input variables, respectively, yielded optimal predictions. Among the level-0 models, the MLPNN model exhibited the highest accuracy, with RMSE = 0.54, MBE = 0.26, MAE = 0.44, R = 0.84, IA = 0.67, and KGE = 0.79. Implementing the stacked ensemble learning model at level-1 significantly improved the SAR prediction compared to the level-0 models. The ensemble-NN model yielded the best performance in estimating SAR within the range of recorded data, with RMSE = 0.53, MBE = 0.29, MAE = 0.41, R = 0.87, IA = 0.70, and KGE = 0.82. Residual analysis further confirmed the superior predictive capability of the level-1 models compared to the level-0 models. The generalized-logistic probability distribution function is used to estimate the extreme values data. The Ensemble-KNN model best predicted extreme values data, with RMSE = 0.69, MBE = −0.61, MAE = 0.61, R = 0.61, IA = 0.26, and KGE = 0.37. The findings underscore the substantial advancements achieved through stacked ensemble methods in enhancing the modeling of SAR across various aspects, including total data, extreme values, and models' residuals.

Variable selection in Poisson regression model based on chaotic meta-heuristic search algorithm

By determining the most significant variables that are connected to the response variable, Increasing prediction accuracy and processing speed can be achieved through the process of variable selection. Regression modeling has drawn a lot of interest from several scientific domains. One of the most effective nature-inspired algorithms that has been suggested recently and can be used effectively for variable selection is the Firefly algorithm. The chaotic firefly algorithm is presented in this work to carry out the Poisson regression model's variable selection. A simulation study is carried out to assess how well the suggested strategy performs in terms of variable selection criteria and prediction accuracy. Its effectiveness is also contrasted with alternative approaches. The outcomes demonstrated the effectiveness of our suggested strategies, which beat other widely used approaches.