Classification And Regression Tree Research Articles

Development and validation of a prediction model for bronchopulmonary dysplasia using respiratory severity score.

To develop and validate a prediction model for severe bronchopulmonary dysplasia (BPD) that integrates the respiratory severity (RS) score with early postnatal risk factors. This retrospective cohort study included preterm infants born at less than 32 weeks gestation or with a birth weight of less than 1500 g, from Aichi Prefecture (training dataset) and Saitama Medical University (validation dataset) from April 1, 2016, to March 31, 2020. The primary outcome was severe BPD, defined as the use of home oxygen therapy or death due to BPD. We used classification and regression tree (CART) analysis to explore the relationship between outcomes and BPD risk factors in the training dataset. The incidence of severe BPD was 149 out of 2026 (7.3%) in the training dataset and 35 out of 387 (8.9%) in the validation dataset. CART analysis identified gestational age and the RS score as significant predictors of outcome in the day 7 and day 14 models, with C-statistics of 0.789 and 0.779, respectively. When applied to the validation dataset, these models achieved C-statistics of 0.753 and 0.827, respectively. Our prediction models demonstrated the ability to predict severe BPD, with the RS score being a crucial predictor. Many existing prediction models for bronchopulmonary dysplasia (BPD) use multiple predictors, and do not provide specific cutoff values, which complicates their clinical application. To address this issue, we developed a prediction model for severe BPD based on a score derived from mean airway pressure and inhaled oxygen concentration at 1-2 weeks of age. This user-friendly model can be easily integrated into clinical practice, facilitating treatment decisions based on predicted probabilities.

Fire Severity Outperforms Remote Sensing Indices in Exploring Post-Fire Vegetation Recovery Dynamics in Complex Plateau Mountainous Regions

Understanding post-fire vegetation recovery dynamics is crucial for damage assessment and recovery planning, yet spatiotemporal patterns in complex plateau environments remain poorly understood. This study addresses this gap by focusing on Yunnan Province, a mountainous plateau region with high fire incidence. We developed an innovative approach combining differenced Normalized Burn Ratio (dNBR) and visual interpretation on Google Earth Engine (GEE) to generate high-quality training samples from Landsat 5 TM/7 ETM+/8 OLI imagery. Four supervised machine learning algorithms were evaluated, with Random Forest (RF) demonstrating superior accuracy (OA = 0.90) for fire severity classification compared to Support Vector Machine (SVM) OA of 0.88, Classification and Regression Tree(CART) OA o f0.85, and Naive Bayes(NB) OA of 0.78. Using RF, we generated annual fire severity maps alongside the Land Surface Water Index (LSWI), Normalized Difference Vegetation Index (NDVI), and Normalized Burn Ratio (NBR) from 2005 to 2020. Key findings include the following: (1) fire severity classification outperformed traditional remote sensing indices in characterizing vegetation recovery; (2) distinct recovery trajectories emerged across severity levels, with moderate areas recovering in 7 years, severe areas transitioning within 2 years, and low severity areas peaking at 2 years post-fire; (3) southern mountainous regions exhibited 1–2 years faster recovery than northern areas. These insights advance understanding of post-fire ecosystem dynamics in complex terrains and support more effective recovery strategies.

Main characteristics of French farms adopting cereal–legume intercropping: A quantitative exploration at the national and local levels

CONTEXTCereal–legume intercropping is a diversification practice that offers many advantages, especially in low-input systems. However, its adoption remains low on European farms, as technical and economic barriers hinder its development. In recent years, an increase in the proportion of arable land cultivated with cereal–legume intercrops has been observed in France. Three areas in particular – in Western, Eastern and Southern France – seem to be particularly dynamic. OBJECTIVEThis study aimed (i) to identify the main farm characteristics associated with the presence of cereal–legume intercrops at the national level in France and (ii) to highlight more specific characteristics that could explain the particular dynamics observed in each focus region. METHODSWe analysed data from the 2020 French Agricultural Census for 43,968 farms representative of the French arable crop, livestock, and mixed crop–livestock farming systems. Through a literature review, we identified key factors linked to the presence of cereal–legume intercrops and related them to 42 variables in the census. At the national level, the most important of these variables were identified and interpreted using a balanced random forest and a classification and regression tree (CART). We tested the CART obtained at the national level in the Western, Eastern, and Southern areas and conducted a random forest analysis for each area to identify local particularities. RESULTS AND CONCLUSIONAt the national level, the presence of cereal–legume intercropping was strongly linked to organic farming and the presence of livestock, especially ruminants. These intercrops were prevalent on farms with high feed autonomy for the cattle and sheep. Additionally, they were commonly observed on farms with grain storage, possibly indicating feed autonomy, on-farm transformation, or marketing outside of agricultural cooperatives. The belonging to a farm machinery cooperative was also strongly associated with cereal–legume intercropping, likely because these cooperatives give farmers access to specific machinery and provide opportunities for knowledge exchange regarding their practices. Similar characteristics were identified at the local level; organic farming was pivotal in the Western and Eastern areas, followed by feed autonomy for cattle. In the Southern area, however, on-farm grain storage capacity was dominant, likely due to longstanding efforts to achieve feed autonomy. SIGNIFICANCEThis exhaustive study on French farms identified key farm characteristics strongly linked to cereal–legume intercrops adoption. This insight is critical for promoting this practice, whether through national public policies or local farming support services. The methodology proposed can be easily reproduced to investigate other farming practices at different spatial scales.

Forest canopy cover estimation with machine learning using GEDI and Landsat data in the Western Marmara Region, Türkiye

Abstract Forests play a crucial role in carbon dioxide absorption, biodiversity preservation, and climate regulation. Timely and precise maps detailing tree cover attributes are vital tools for the fields of environmental research and natural resource management. This study addresses the challenge of accurately estimating forest canopy cover by integrating Global Ecosystem Dynamics Investigation (GEDI) and Landsat data in the Eastern Marmara Region of Türkiye. Despite the advancements in remote sensing technologies such as light detection and ranging (LiDAR) and optical sensors, and the importance of forest canopy cover in assessing forest health and carbon stocks, there is limited research on utilizing spaceborne GEDI Level 2B data for forest canopy cover mapping. Six different machine learning methods were employed, namely Classification and Regression Trees (CART), Categorical Boosting (CTB), Light Gradient Boosting Machines (LGBM), Multilayer Perceptron (MLP), Random Forest (RF), and Extreme Gradient Boosting (XGB), to generate forest canopy cover maps using Landsat 8 and 9 satellite images with a variety of vegetation indices and texture features. Model performances were evaluated using metrics such as R2, Root Mean Square Error (RMSE), and Median Absolute Error (MdAE), with statistical significance assessed via Friedman and Wilcoxon signed-rank tests. The results of the tests indicated that the XGB (R2 = 0.5570, RMSE = 0.1603, MdAE = 0.0885) and RF (R2 = 0.5497, RMSE = 0.1617, MdAE = 0.0896) algorithms, which were trained with GEDI Level 2B data, provided greater accuracy in forest canopy cover estimation compared to the other algorithms. This study offers insight into the prediction performance of GEDI Level 2B spaceborne LiDAR data in conjunction with XGB and RF algorithms for forest canopy cover estimation and underscores the significance of integrating advanced remote sensing data for forest monitoring.

P-1270. Using Machine Learning and ICD-10 Codes to Identify Cases of Laboratory-Confirmed Influenza, 2015-2020, US Flu VE Network

Abstract Background The International Classification of Diseases (ICD-10) has unique codes for influenza. Accurate classification of influenza is necessary to calculate unbiased vaccine effectiveness and typically requires laboratory testing. We used the US Influenza Vaccine Effectiveness (Flu VE) Network to determine if outpatient influenza-positive cases and influenza-negative controls could be identified by the presence or absence of certain codes to better understand the role of ICD-10 in estimating VE.Table:Twelve most common ICD-10 codes among influenza-positive outpatients in the US Flu VE Network and their corresponding rankings and frequencies among influenza-negative patients, 2015-16 through 2019-20 influenza seasons Methods During the 2015-16 through 2019-20 influenza seasons, 5 US Flu VE sites enrolled outpatients aged ≥6 months with medically attended acute respiratory illness (MAARI) defined as self-reported new or worsening cough lasting ≤7 days and tested patients for influenza by RT-PCR. We determined the frequencies of ICD-10 codes among influenza cases and controls. We also used classification and regression trees (CART) models to evaluate ICD-10 codes as predictors of influenza test results.Figure:Classification and Regression Tree using ICD-10 codes to determine influenza testing status for outpatients in the US Flu VE Network, 2015-16 through 2019-20 influenza seasons Results Over 5 seasons in the US Flu VE, we identified 2,711 ICD-10 codes assigned to enrollments among MAARI patients. Among 11,861 patients with laboratory-confirmed influenza, we identified a total of 1,234 unique ICD-10 codes versus 2,450 ICD-10 codes among 30,622 patients who tested negative for influenza. Codes for cases and controls overlapped substantially, with Diseases of the Respiratory system (J00-J99) being the most frequent (78%) among both. Among cases and controls, J06.9 (Acute upper respiratory infection, unspecified) and R05 (Cough) were the most frequent (Table). CART analysis identified a subset of 6 optimal MAARI-associated ICD-10 codes that identified 6,112 (52%) influenza cases, and these codes were absent for 25,518 (83%) test-negative controls totaling 31,630 (74%) cases and controls (Figure A). Conclusion Our findings show the importance of laboratory-confirmed influenza outcomes for influenza VE analyses; machine learning may be useful in identifying MAARI-associated ICD-10 diagnostic codes to define patients for inclusion in test-negative VE studies. In our study, we identified a restricted list of ICD-10 codes that could be used to improve VE estimates in electronic medical record systems with access to testing data. Disclosures Huong Nguyen, PhD, MPH, CSL Seqirus: Advisor/Consultant|CSL Seqirus: Grant/Research Support|GSK: Grant/Research Support|ModernaTX: Advisor/Consultant|ModernaTX: Grant/Research Support Mary Patricia Nowalk, PhD, AstraZeneca - Icosavax: Grant/Research Support|GSK: Advisor/Consultant|Merck & Co.: Grant/Research Support|Sanofi: Grant/Research Support Arnold Monto, MD, Roche: Advisor/Consultant

P-1337. Analyzing SARS-CoV-2 Case Numbers and Clustering to Predict a Nursing Home Outbreak

Abstract Background Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2), can spread rapidly in nursing homes (NH). Intervening to reduce spread should follow a signal predicting an imminent "outbreak". Various thresholds for declaring a COVID-19 outbreak have considered local factors, viral variants, case identification timelines and vaccination rates. We evaluated the number and interval between initial SARS-CoV-2 infections (cases) most predictive of future cases using VA Community Living Centers (CLCs) as NH analogues. Methods We captured SARS-CoV-2 PCR test results from individual NH neighborhoods (specialized wards). After an initial index case, we evaluated for additional cases over the next 28 days by R² analysis and 14 days by Classification and Regression Trees (CART) analysis. We evaluated the R2 goodness-of-fit measure from separate linear models by neighborhood to measure the predictive power of various outbreak definitions, adjusted for the number of residents in the neighborhood. Also, we used CART creating a tree-like structure to predict continuous target variables to see the effect of census and time at only 14 days. Results We included 136 CLCs housing 16,339 Veterans, mean age of 74, mostly male (96%) and white (68%), in 490 neighborhoods, representing 1,252,037 Veteran exposure days from 12/18/2021 to 06/18/2022, and 1,765 SARS-CoV-2 infections. We divided neighborhoods by small [< 10 Veterans (n=30,390)], medium [10-15 veterans (n=25,349)], and large [ >15 veterans (n=31,899)]. The most predictive outbreak definition for future COVID-19 cases over the next 14 days for our linear model analysis was: 2 cases across all neighborhoods, 4 cases for large, 2 for medium and 1 for small size neighborhoods. Our preliminary CART analysis suggested the first case was predictive of another case in 5 days, but if ≥4 occurred in 5 consecutive days, it predicted a large-scale outbreak with 10 or more cases over the next 14 days if the neighborhood is large enough to have that many cases. Conclusion Our analysis suggests when 4 SARS-CoV-2 cases cluster in five days in a large neighborhood, a large outbreak is imminent, potentially warranting aggressive control measures. However further refinement of our methods is needed to confirm this conclusion. Disclosures Yasin Abul, MD, Moderna: Grant/Research Support|Moderna, Abt, CDC: Grant/Research Support Kevin McConeghy, Pharm.D., Genentech: Grant/Research Support|GlaxoSmithKline: Grant/Research Support|Moderna: Grant/Research Support|Sanofi-Pasteur: Grant/Research Support|Seqirus: Grant/Research Support Stefan Gravenstein, MD, MPH, CDC: Advisor/Consultant|CDC: Grant/Research Support|Genentech: Advisor/Consultant|Genentech: Grant/Research Support|Genentech: Honoraria|GlaxoSmithKline: Advisor/Consultant|GlaxoSmithKline: Grant/Research Support|GlaxoSmithKline: Honoraria|Janssen: Advisor/Consultant|Janssen: Grant/Research Support|Janssen: Honoraria|Moderna: Advisor/Consultant|Moderna: Grant/Research Support|Moderna: Honoraria|NIH: Grant/Research Support|Pfizer: Advisor/Consultant|Pfizer: Grant/Research Support|Pfizer: Honoraria|Sanofi: Advisor/Consultant|Sanofi: Grant/Research Support|Sanofi: Honoraria|Seqirus: Advisor/Consultant

Optimal selection of satellite XCO2 images for urban CO2 emission monitoring

Abstract. There is a growing interest in estimating urban CO2 emission from spaceborne imagery of the CO2 column-average dry-air mole fraction (XCO2). Emission estimation methods have been widely tested and applied to actual or synthetic images. However, there is still a lack of objective criteria for selecting images that are worth processing. This study analyzes the performances of an automated method for estimating urban emissions as a function of the targeted cities and of the atmospheric conditions. It uses synthetic data experiments with synthetic truth and 9920 synthetic satellite images of XCO2 over 31 of the largest cities across the world generated with a global adaptive-mesh model, the Ocean–Land–Atmosphere Model (OLAM), zoomed in at high resolution over these cities. We use a decision tree learning method applied to this ensemble of synthetic images to define criteria based on these emission and atmospheric conditions for the selection of suitable satellite images. We show that our automated method for the emission estimation, based on a Gaussian plume model, manages to produce estimates for 92 % of the synthetic images. Our learning method identifies two criteria, the wind direction's spatial variability and the targeted city's emission budget, that discriminate images whose processing yields reasonable emission estimates from those whose processing yields large errors. Images corresponding to low spatial variability in wind direction (less than 12°) and to high urban emissions (greater than 2.1 kt CO2 h−1) account for 47 % of the images, and their processing yields relative errors in the emission estimates with a median value of −7 % and an interquartile range (IQR) of 56 %. Images corresponding to a high spatial variability in wind direction or to low urban emissions account for 53 % of our images, and their processing yield relative errors in the emission estimates with a median value of −31 % and an IQR of 99 %. Despite such efficient filtering, the accuracy of the estimates corresponding to the former group of images varies widely from city to city.

Efficient diagnosis of diabetes mellitus using an improved ensemble method

Diabetes is a growing health concern in developing countries, causing considerable mortality rates. While machine learning (ML) approaches have been widely used to improve early detection and treatment, several studies have shown low classification accuracies due to overfitting, underfitting, and data noise. This research employs parallel and sequential ensemble ML approaches paired with feature selection techniques to boost classification accuracy. The Pima India Diabetes Data from the UCI ML Repository served as the dataset. Data preprocessing included cleaning the dataset by replacing missing values with column means and selecting highly correlated features using forward and backward selection methods. The dataset was split into two parts: training (70%), and testing (30%). Python was used for classification in Jupyter Notebook, and there were two design phases. The first phase utilized J48, Classification and Regression Tree (CART), and Decision Stump (DS) to create a random forest model. The second phase employed the same algorithms alongside sequential ensemble methods—XG Boost, AdaBoostM1, and Gradient Boosting—using an average voting algorithm for binary classification. Evaluation revealed that XG Boost, AdaBoostM1, and Gradient Boosting achieved classification accuracies of 100%, with performance metrics including F1 score, MCC, Precision, Recall, AUC-ROC, and AUC-PR all equal to 1.00, indicating reliable predictions of diabetes presence. Researchers and practitioners can leverage the predictive model developed in this work to make quick predictions of diabetes mellitus, which could save many lives.

Short-Wave Infrared Spectroscopy for On-Site Discrimination of Hazardous Mineral Fibers Using Machine Learning Techniques

Asbestos fibers are well-known carcinogens, and their rapid detection is critical for ensuring safety, protecting public health, and promoting environmental sustainability. In this work, short-wave infrared (SWIR) spectroscopy, combined with machine learning (ML), was evaluated as an environmentally friendly analytical approach for simultaneously distinguishing the asbestos type, asbestos-containing materials in various forms, asbestos-contaminated/-uncontaminated soil, and asbestos-contaminated/-uncontaminated cement, simultaneously. This approach offers a noninvasive and efficient alternative to traditional laboratory methods, aligning with sustainable practices by reducing hazardous waste generation and enabling in situ testing. Different chemometrics techniques were applied to discriminate the material classes. In more detail, partial least squares discriminant analysis (PLS-DA), principal component analysis-based discriminant analysis (PCA-DA), principal component analysis-based K-nearest neighbors classification (PCA-KNN), classification and regression trees (CART), and error-correcting output-coding support vector machine (ECOC SVM) classifiers were tested. The tested classifiers showed different performances in discriminating between the analyzed samples. CART and ECOC SVM performed best (RecallM and AccuracyM equal to 1.00), followed by PCA-KNN (RecallM of 0.98–1.00 and AccuracyM equal to 1.00). Poorer performances were obtained by PLS-DA (RecallM of 0.68–0.72 and AccuracyM equal to 0.95) and PCA-DA (RecallM of 0.66–0.70 and AccuracyM equal to 0.95). This research aligns with the United Nations’ Sustainable Development Goals (SDGs), particularly SDG 3 (Good Health and Well-Being), by enhancing human health protection through advanced asbestos detection methods, and SDG 12 (Responsible Consumption and Production), by promoting sustainable, low-waste testing methodologies.

Preoperative and early postoperative factors associated with visual outcome after macular-off retinal detachment surgery.

To evaluate preoperative and early postoperative factors associated with changes in visual acuity in macula-off rhegmatogenous retinal detachment (RRD) patients. A retrospective review of 59 eyes from 59 patients who underwent pars plana vitrectomy (PPV) and silicone oil (SiO) tamponade for macula-off RRD. Preoperative and early postoperative characteristics were analyzed using the classification and regression tree (CART) analysis. The baseline preoperative BCVA improved from 1.67 ± 0.97 logMAR to 0.21 ± 0.22 logMAR at final visit. Central foveal thickness (CFT) and BCVA showed significant increases from 1 week post-surgery onward. Mean CFT significantly increased from 136.97 ± 40.17 μm to 192.54 ± 34.09 μm (p < 0.001). The percentage of eyes with integrated external limiting membrane (ELM) and ellipsoid zone (EZ) was progressively increased from 45.76% and 10.17% to 98.31% and 71.19%, respectively (p < 0.001). BCVA at 1week after surgery, ELM integrity, preoperative intraretinal cysts, and macula-off duration were significantly associated with final visual recovery after surgery. The first week postoperative BCVA is the most powerful predictor for final BCVA. ELM integrity, intraretinal cysts, and macula-off duration are also significantly associated with final visual recovery.

Identifying Depression Using Machine Learning

Depression is the leading cause of disability worldwide. However, accurately estimating the epidemiological factors that contribute to depression remains challenging. Deep learning algorithms can be used to assess the factors that contribute to the prevalence and clinical manifestations of depression. In this paper, five machine learning models, logistic regression, decision tree, random forest, SVM, and CatBoost, were used to assess depression in 5533 adult participants from the 2018 NHANES database. The results show that random forest is the best model for identifying depression, with the highest area under the working characteristic curve (AUC), followed by CatBoost and decision tree models. This suggests that using machine learning can accurately predict depression risk.

Amino acids and CART distinguish A-β+ Ketosis-Prone Diabetes from type 1 and type 2 diabetes during hyperglycemic crises.

When clinically stable, patients with A-β+ Ketosis-Prone Diabetes (KPD) manifest unique markers of amino acid metabolism. Biomarkers differentiating KPD from type 1 (T1D) and type 2 diabetes (T2D) during hyperglycemic crises would accelerate diagnosis and management. Compare serum metabolomics of KPD, T1D and T2D patients during hyperglycemic crises, and utilize Classification and Regression Tree (CART) modeling to distinguish these forms of diabetes. Urban hospital emergency center. Adults with KPD, T1D and T2D during hyperglycemic crises (with or without diabetic ketoacidosis (DKA), and healthy controls. Comparisons of serum metabolite and hormonal profiles, and CART analysis. Group differences in concentrations of amino acids, their metabolites and relationship to glucose counterregulatory hormones; C-peptide cutoffs and analytes to distinguish KPD, T1D and T2D. Concentrations of most amino acids were similar in KPD and T1D and lower compared to T2D (P<0.05). Glucagon and cortisol concentrations were correlated with 3-methylhistidine and blood urea nitrogen in KPD but not in T1D. A C-peptide cutoff of 0.496 ng/mL differentiated T1D from KPD during DKA. CART revealed that a regression model based on the concentrations of β-hydroxybutyrate, C-peptide, glucagon, alpha-keto-β-methylvalerate, cystine and myristoyl-L-carnitine distinguished KPD from T1D and T2D. During DKA, KPD and T1D patients have similarly altered amino acid profiles that differentiate them from T2D patients. Elevated protein catabolic hormones drive altered amino acid metabolism in KPD, rather than insulin deficiency as with T1D. A combination of 6 analytes differentiates KPD from T1D and T2D during hyperglycemic crises.

Small and medium-scale biorefineries: biomass quantification and its bioeconomic potential in the Southern Coastal Territory of Bahia.

Small and medium-scale biorefineries are processing facilities designed to produce a portfolio of value-added products with applications in different productive sectors. The Territory of the Southern Coast (TSC) of Bahia in Brazil has a high potential to provide agricultural and forest biomass for these bioindustries. This study focuses on quantifying the biomass of key agricultural crops in the TSC from 1999 to 2019, utilizing data from the Municipal Agricultural Production (MAP) survey conducted by the Brazilian Institute of Geography and Statistics (IBGE). The biomass is subsequently categorized based on its applications in pharmaceutical and other industries, employing the Classification and Regression Tree (CART) algorithm, specifically using the Gini Index. Thus, our results reveal a noteworthy trend over 20years through a time series analysis. For example, banana production (1.3 Mt) surpassed cocoa production (1.1 Mt) despite cocoa cultivation covering a vast 5.8 million hectares expanse compared to the smaller banana footprint. Cocoa and cassava are significant contributors, representing 41.95% and 30.51% of the total raw material, collectively constituting 72.46% of the TSC biomass. In addition, coffee biomass has been profiled as a raw material with potential applications across various industries, while bananas are oriented towards supplying pharmaceutical inputs. Forest species also have the potential for use in the pharmaceutical industry. In conclusion, small and medium-scale biorefineries can help increase agricultural and forest biomass use to provide inputs for various industries and guarantee positive social impacts for the population.

A Novel Method to Forecast Nitrate Concentration Levels in Irrigation Areas for Sustainable Agriculture

This study developed an ANN-based model to predict nitrate concentrations in drainage waters using parameters that are simpler and more cost-effective to measure within the Lower Seyhan Basin, a key agricultural region in Turkey. For this purpose, daily water samples were collected from a drainage measurement station during the 2022 and 2023 water years, and nitrate concentrations were determined in the laboratory. In addition to nitrate concentrations, other parameters, such as flow rate, EC, pH, and precipitation, were also measured simultaneously. The complex relationship between measured nitrate values and other parameters, which are easier and less costly to measure, was used in two different scenarios during the training phase of the ANN-Nitrate model. After the model was trained, nitrate values were estimated for the two scenarios using only the other parameters. In Scenario I, random values from the dataset were predicted, while in Scenario II, predictions were made as a time series, and model results were compared with measured values for both scenarios. The proposed model reliably fills dataset gaps (Scenario I) and predicts nitrate values in time series (Scenario II). The proposed model, although based on an artificial neural network (ANN), also has the potential to be adapted for methods used in machine learning and artificial intelligence, such as Support Vector Machines, Decision Trees, Random Forests, and Ensemble Learning Methods.

Use of data mining algorithms in prediction of eggshell thickness from egg quality traits of Potchefstroom Koekoek layers

Egg quality is affected by lot of factors. Study was conducted to compare performance of data mining algorithms; Classification and regression tree (CART), Chi-square automatic interaction detection (CHAID), Exhaustive chi-square automatic interaction detection (Ex-CHAID) and Multivariate adaptive regression spline (MARS) in prediction of Potchefstroom Koekoek’s eggshell thickness from egg quality traits. 350 eggs were collected at 31st to 39th week to examine the egg quality traits. MARS with R2(0.86) revealed yolk ratio, shell weight, egg shape index, yolk ratio, shell ratio, albumen weight and albumen ratio as explanatory variables predicting eggshell thickness. CART with R2 (0.37), yolk/albumen ratio was noted to be influential predictor of eggshell thickness. CHAID and Ex-CHAID (R2= 0.35) discovered egg weight as the best predictor of eggshell thickness. MARS with R2(0.86) revealed yolk ratio, shell weight, egg shape index, yolk ratio, shell ratio, albumen weight and albumen ratio as explanatory variables predicting eggshell thickness. MARS had high r (0.925), R2 (0.856) and lower RMSE (0.129) and AIC (-975.331) compared to CHAID, Ex-CHAID and CART leading MARS to be the best data mining algorithm when predicting the eggshell thickness using egg quality traits.

Artificial Intelligence Technologies in Predicting Life Insurance Premiums: A Case Study on the Iraqi General Insurance Company

The insurance sector relies on considerable amounts of data and statistics where standard actuarial computations form the basis for determining risks, liabilities, and appropriate premiums. The traditional methods make predicting life insurance premiums complex task and less efficient. This affects the overall cost and development of proper insurance solutions. However, the rapid development of artificial intelligence technologies presents great opportunities to reshape traditional processes to improve efficiency and accuracy within the insurance industry. The presented work focuses on applying artificial intelligence in developing actuarial calculations, aiming to find a better estimate of the life insurance premium and providing better quality reports for the Iraqi General Insurance Company. In the current study, we used three machine learning algorithms, namely XGBoost, Random Forest (RF), and Decision Tree (DT), to analyze the collected data and estimate the life insurance premium. Different regression evaluation metrics, including MAE, RMSE, R², and Adjusted R², were applied in testing the performance of the algorithms. The results from the experiments showed that the XGBoost algorithm was the highest performing compared to other algorithms that recorded the lowest MAE and the highest R² score. The research also suggested that incorporating artificial intelligence techniques into actuarial analysis can enhance financial reporting efficiency, thus increasing insurance companies' competitiveness in the important sector. Consequently, Iraqi insurance companies need to adopt such technologies to face the changing insurance conditions by training employees and adopting digital transformation to stay competitive and benefit from innovation.

Improving groundwater quality predictions in semi-arid regions using ensemble learning models.

Groundwater resources constitute one of the primary sources of freshwater in semi-arid and arid climates. Monitoring the groundwater quality is an essential component of environmental management. In this study, a comprehensive comparison was conducted to analyze the performance of nine ensembles and regular machine learning (ML) methods in predicting two water quality parameters including total dissolved solids (TDS) and pH, in an area with semi-arid climate conditions. The study area under consideration is an aquifer located in the Sirjan plain, Kerman, Iran. The developed models include standard multilayer perceptron neural network (MLPNN), classification and regression trees (CART), Chi-square automatic interaction detection (CHAID), and their ensemble versions in bagging (BG) and boosting (BT) ensemble structures. The analysis revealed that standard MLs yield comparable results in predicting TDS. The MLPNN, exhibiting a standard root mean square error (SRMSE) of 0.085, demonstrated superior accuracy in predicting TDS when contrasted with CART and CHAID models. Predicting pH poses a greater challenge for the models. Ensemble techniques significantly enhanced the accuracy of regular models. On average, the bagging and boosting techniques resulted in a 22.68% improvement in the accuracy of regular models, which represents a statistically significant enhancement. The boosting method, with an average SRMSE of 0.0602, is more accurate than bagging. Based on the results, the CHAID-BT with SRMSE of 0.0790 and CHAID-BG with SRMSE of 0.0330 are ranked the most accurate models for predicting TDS and pH, respectively. The performance of ensemble techniques in predicting TDS is more remarkable. In practical implementation, ensemble techniques can be considered an alternative method with high accuracy for sustainable water resources management in semi-arid regions, helping to address water shortages, climate change, water pollution, etc.

Classification and Regression Trees analysis identifies patients at high risk for kidney function decline following hospitalization.

Estimated glomerular filtration rate (eGFR) decline is associated with negative health outcomes, but the use of decision tree algorithms to predict eGFR decline is underreported. Among patients hospitalized during the first year of the COVID-19 pandemic, it remains unclear which individuals are at the greatest risk of eGFR decline after discharge. We conducted a retrospective cohort study on patients hospitalized at Stony Brook University Hospital in 2020 who were followed for 36 months post discharge. Random Forest (RF) identified the top ten features associated with fast eGFR decline. Logistic regression (LR) and Classification and Regression Trees (CART) were then employed to uncover the relative importance of these top features and identify the highest risk patients. In the cohort of 1,747 hospital survivors, 61.6% experienced fast eGFR decline, which was associated with younger age, higher baseline eGFR, and acute kidney injury (AKI). Multivariate LR analysis showed that older age was associated with lower odds of fast eGFR decline whereas length of hospitalization and vasopressor use with greater odds. CART analysis identified length of hospitalization as the most important factor and that patients with AKI and hospitalization of 27 days or more were at highest risk. After grouping by ICU and COVID-19 status and propensity score matching for demographics, these risk factors of fast eGFR decline remained consistent. CART analysis can help identify patient subgroups with the highest risk of post-discharge eGFR decline. Clinicians should consider the length of hospitalization in post-discharge monitoring of kidney function.

Review and practical excursus on the Comparison between traditional statics methods and Classification And Regression Tree (CART) in real-life data: Low protein diet compared to Mediterranean diet in patients with Chronic Kidney Disease

Review and practical excursus on the Comparison between traditional statics methods and Classification And Regression Tree (CART) in real-life data: Low protein diet compared to Mediterranean diet in patients with Chronic Kidney Disease

Automated mold defects classification in paintings: A comparison of machine learning and rule-based techniques.

Mold defects pose a significant risk to the preservation of valuable fine art paintings, typically arising from fungal growth in humid environments. This paper presents a novel approach for detecting and categorizing mold defects in fine art paintings. The technique leverages a feature extraction method called Derivative Level Thresholding to pinpoint suspicious regions within an image. Subsequently, these regions are classified as mold defects using either morphological filtering or machine learning models such as Classification and Regression Trees (CART) and Linear Discriminant Analysis (LDA). The efficacy of these methods was evaluated using the Mold Features Dataset (MFD) and a separate set of test images. Results indicate that both methods improve the accuracy and precision of mold defect detection compared to no classifier. However, the CART algorithm exhibits superior performance, increasing precision by 32% to 53% while maintaining high accuracy (96%) even with an imbalanced dataset. This innovative method has the potential to transform the approach to managing mold defects in fine art paintings by offering a more precise and efficient means of identification. By enabling early detection of mold defects, this method can play a crucial role in safeguarding these invaluable artworks for future generations.